Digital biomarkers are healthcare’s next frontier

Blood pressure, body temperature, hemoglobin A1c levels and other biomarkers have been used for decades to track disease. While this information is essential for chronic condition management, these and many other physiological measurements are typically captured only periodically, making it difficult to reliably detect early meaningful changes.

Moreover, biomarkers extracted from blood require uncomfortable blood draws, can be expensive to analyze, and again, are not always timely.

Historically, continuous tracking of an individual’s vital signs meant they had to be in a hospital. But that’s not true anymore. Digital biomarkers, collected from wearable sensors or through a device, offer healthcare providers an abundance of traditional and new data to precisely monitor and even predict a patient’s disease trajectory.

With cloud-based servers and sophisticated, yet inexpensive, sensors both on the body and off, patients can be monitored at home more effectively than in a hospital, especially when the sensor data is analyzed with artificial intelligence (AI) and machine-learning technology.

Opportunities for digital biomarkers

A major opportunity for digital biomarkers is in addressing neurodegenerative diseases such as mild cognitive impairment, Alzheimer’s disease and Parkinson’s disease.

Neurodegenerative disease is a major target for digital biomarker development due to a lack of easily accessible indicators that can help providers diagnose and manage these conditions. A definitive diagnosis for Alzheimer’s disease today, for example, generally requires positron emission tomography (PET), magnetic resonance imaging (MRI) or other imaging studies, which are often expensive and not always accurate or reliable.

Cost savings and other benefits

Digital biomarkers have the potential to unlock significant value for healthcare providers, companies and, most importantly, patients and families, by detecting and slowing the development of these diseases.

Endel’s generative soundscapes show up in Sony’s new headphones

The other day, Brian reported on Sony’s new LinkBuds headphones, including its partnership with “what if Brian Eno was a piece of computer software” app Endel. The company uses really fascinating AI technology to generate soundscapes and music tracks to help your brain do its best work — to help you focus deeper, sleep more easily or to relax you. I spoke with one of Endel’s founders to learn more about the tech and its deal with Sony.

“Endel is first and foremost a technology that was built to help you focus, relax and sleep. And the way this technology works, it procedurally generates a soundscape in real time on the spot, on the device. It is personalized to you based on a number of inputs that we collect about you; things like the time of day, your heart rate, the weather, your movement and your circadian rhythms, like how much sleep you got last night,” explains Oleg Stavitsky, CEO and co-founder at Endel. “This technology listens to all of this data, plugs into the algorithm, which creates the soundscape in real time, which allows us to react in real time to the changes in you. Using this technology, we are building an ecosystem of products, so that our soundscapes can follow you everywhere during the day across all these channels and platforms. We are pretty much everywhere at this point; iOS, Android, Apple Watch, Mac or Apple TV, Alexa… you name it.”

In reviewing the product I did stumble across a couple of glaring omissions in where it is available: There was no way of streaming it to my Sonos speakers (the workaround is to install Alexa on Sonos), and the Endel app doesn’t support casting, so you can’t stream to Google Home either.

Running the app using earphones, however, creates an intimate and beautiful experience. The audio tracks are Eno-esque in their expansiveness; it’s like a slowly evolving ambient soundtrack to your day. Sitting at my desk, I felt myself focus; a combination of the music and blocking and drowning out distractions.

The soundscapes are stem-based — professional music industry jargon for snippets of sounds, think of them as samples. The app has a huge library of samples and stems, and the algorithm picks the right stems to sequence the audio together. On top of the basic sequencing, the software runs additional adjustments on top.

“We have a few AI systems on top of that sequencer; AI systems that generate melodies basically. There are millions and millions and millions of variations,” says Stavitsky. “Some of the soundscapes on the app are done in collaboration with some of the biggest artists on the planet. We have Grimes and Miguel and James Blake and Plastic Man and others that we’ve worked with, so they are good. The way they work with us is they prepare a stem pack, a sound pack. They never submit a musical composition. They just are the building blocks that the algorithm then uses to assemble tracks on the fly.”

The company says it’s approached by companies all the time, and have to consider whether partnerships are a cost or a benefit at any given time. It decided to say “yes” to headphones giant Sony, resulting in this collaboration.

“Sony’s headphones innovation department approached us. They said we’re working on this new model that will somehow understand the context of where you are, and we want those headphones to proactively activate a certain soundscape,” says Stavitsky, “I’m frankly very, very skeptical about all these integrations, for a number of reasons. There’s always an opportunity cost. Being a small company, you’re wondering if we should do this. What got me excited about this is that the fundamental idea of Endel is that it’s an always-on soundscape that follows you everywhere during the day. Sometimes you can barely hear it, and sometimes it’s like front and center and it shields you from the rest of the world. I think this idea of headphones that proactively trigger a certain kind of soundscape depending on the context of what’s happening with you is exactly how we envision how our product is used. This is just going to be one huge play button — you press that button, and it listens to your calendar, listens to your heart rate, and it proactively shifts between all of the soundscapes. That’s what we are working toward, and these headphones make that real.”

AI’s role is poised to change monumentally in 2022 and beyond

The latest developments in technology make it clear that we are on the precipice of a monumental shift in how artificial intelligence (AI) is employed in our lives and businesses.

First, let me address the misconception that AI is synonymous with algorithms and automation. This misconception exists because of marketing. Think about it: When was the last time you previewed a new SaaS or tech product that wasn’t “fueled by” AI? This term is becoming something like “all-natural” on food packaging: ever-present and practically meaningless.

Real AI, however, is foundational to supporting the future of how businesses and individuals function in the world, and a huge advance in AI frameworks is accelerating progress.

As a product manager in the deep learning space, I know that current commercial and business uses of AI don’t come close to representing its full or future potential. In fact, I contend that we’ve only scratched the surface.

Ambient computing

The next generation of AI products will extend the applications for ambient computing.

  • Ambient = in your environment.
  • Computing = computational processes.

We’ve all grown accustomed to asking Siri for directions or having Alexa manage our calendar notifications, and these systems can also be used to automate tasks or settings. That is probably the most accessible illustration of a form of ambient computing.

Ambient computing involves a device performing tasks without direct commands — hence the “ambient,” or the concept of it being “in the background.” In ambient computing, the gap between human intelligence and artificial intelligence narrows considerably. Some of the technologies used to achieve this include motion tracking, wearables, speech-recognition software and gesture recognition. All of this serves to create an experience in which humans wish and machines execute.

The Internet of Things (IoT) has unlocked continuous connectivity and data transference, meaning devices and systems can communicate with each other. With a network of connected devices, it’s easy to envision a future in which human experiences are effortlessly supported by machines at every turn.

But ambient computing is not nearly as useful without AI, which provides the patterning, helping software “learn” the norms and trends well enough to anticipate our routines and accomplish tasks that support our daily lives.

On an individual level, this is interesting and makes life easier. But as professionals and entrepreneurs, it’s important to see the broader market realities of how ambient computing and AI will support future innovation.

Announcing the early-stage startups exhibiting at TC Sessions: Climate 2022

Examining the role that technology and startups play in fighting climate change is the raison d’être for TC Sessions: Climate 2022 (featuring the Extreme Tech Challenge 2022 Global Finals), which takes place on June 14 at UC Berkeley (with an online day June 16).

Pro tip: Buy your pass now, and you’ll save $200. Prices go up at the door.

One of our favorite aspects of this event takes place on the exhibition floor. That’s where you’ll meet, connect — and maybe even find opportunities to collaborate — with early-stage startups showcasing a wide range of climate tech solutions. Based on the companies listed below, it promises to be an impressive display of innovative products and solutions designed to help humanity mitigate and adapt to climate change. 

And don’t miss the interviews, presentations and breakouts with the leading minds and makers who are dedicated to creating or investing in saving our planet and its inhabitants. Check out the event agenda and plan your strategy.

Alrighty then and without further ado, here are the early-stage climate tech startups exhibiting at TC Sessions: Climate 2022.

Algeon Materials: Algeon Materials is on a mission to fight climate change and reduce plastic pollution by creating sustainable and biodegradable plastic alternatives from kelp. 

Arloid Automation: SaaS for energy optimization in the real estate sector. Through precise and proactive management of individual HVAC (heating, ventilation, a/c) devices at a granular level, our solution instantly delivers up to 40% reduction in utilities consumption and carbon footprint. We operate a gain-share model that requires no capital expenditure and no upfront payments from our customers. 

Atmo: Atmo builds AI-based climate computers for governments and enterprises. 

ChargeNet Stations: ChargeNet democratizes electric vehicle access, makes charging clean and convenient, prevents greenhouse gas emissions and transforms restaurant parking lots into profit centers. 

Cincel: The multi-signature, notarization, blockchain and identity suite that streamlines the life cycle of signing documents with legal and technological compliance. 

Corigin Solutions: A carbon-removal company that converts crop waste into regenerative agricultural inputs and carbon sequestration solutions. The first plant is up and running in Merced, California. 

DiviGas: A polymeric hydrogen separation membrane for H2 separation from CO2 and other harsh gasses. Powers gray/blue/green hydrogen use cases and billions of dollars in industrial products.

Earthshot Labs: Planetary-scale ecological regeneration. 

FarmSense: Climate change affects pest impact on crops. FarmSense provides real-time autonomous pest monitoring that reduces labor and unnecessary broad-spectrum pesticide sprays. 

Nanospan: Innovates in disruptive, decarbonizing material technologies aimed at sustainable living, for the construction tech and energy storage sectors.

NovoNutrients: NovoNutrients will capture hundreds of megatons of carbon and feed the world. We are the biotech company that’s cleanly turning hydrogen and industrial carbon dioxide emissions into microbial, alternative protein ingredients. 

Orbillion Bio: Cell-cultured premium meat. Working with butchers and chefs, our proprietary tech lets them develop four different heritage meats 18 times faster and 10 times cheaper than competitors. 

Paces: Actionable data insights to help green infrastructure developers, operators and investors to build more projects. 

Parcel Health: Creates sustainable medication packaging for the pharmaceutical industry.

Plantd: Pioneering the low-emissions factory of the future to transform rapidly renewable grass into durable, carbon-negative building materials. 

PSapling Solar: Superior substitute to roof-top solar, capable of scaling residential and business solar so fast that utility companies and governments are pressed to accelerate their own green footprint, mitigating climate change.

SeekInvest: Values-based financial services platform. 

Solutum: A novel plastic material — with all the functionality and benefits of traditional plastic — that completely dissolves in water at ambient temperature. After a predetermined time delay, it bio-degrades to natural and eco-friendly components with no microplastic residue. 

Sphere: Sphere makes it easy for employers to offer authentic climate-friendly investment options in their 401(k) retirement plans. 

Undesert Corporation: Capturing carbon, 100% driven with renewable solar energy; curating forests; cleaning water; afforesting deserts. 

WeCantWait.World: To respond to urgent causes, advance essential solutions, and invest for a flourishing future, we will consider giving away a lot [of money] to bold and effective organizations. 

TC Sessions: Climate 2022 takes place at UC Berkley’s Zellerbach Auditorium in Berkeley on June 14 with an online event on June 16. Don’t miss your opportunity to meet, connect and collaborate with some of the leading early-stage climate tech startups. Buy your pass today and save $200.

Is your company interested in sponsoring or exhibiting at TC Sessions Climate 2022? Contact our sponsorship sales team by filling out this form.

Data intelligence startup Near, with 1.6B anonymized user IDs, lists on Nasdaq via SPAC at a $1B market cap; raises $100M

The IPO window has all but closed for technology companies in the wake of a massive downturn in the market, but an opening still remains for some, in the form of SPACs. Near — a data intelligence company that has amassed 1.6 billion anonymized user profiles attached to 70 million locations in 44 countries — today announced that it would be listing on Nasdaq by way of a merger with KludeIn I Acquisition Corp., one of the many blank check companies that have been set up for the purposes of taking privately held companies public, at a valuation “near” $1 billion. It will trade on Nasdaq using the “NIR” ticker.

Alongside that, the company is picking up a $100 million equity investment into its business from CF Principal Investments, an affiliate of Cantor Fitzgerald. 

If you’ve been following Near or the SPAC market, you might recall that there were rumors of KludeIn talking to Near back in December. At the time Near was reportedly aiming at a valuation of between $1 billion and $1.2 billion with the listing. The last several months, however, have seen the IPO market virtually shut down alongside a massive drop in technology stocks across the board and a wider downturn in tech investing overall, even in much smaller, earlier-stage startups.

Near, originally founded in Singapore in 2012 and now based out of Pasadena, had raised around $134 million in funding, including a $100 million round in 2019 — which had been the company’s last big raise.

Its investors include the likes of Sequoia India, JP Morgan, Cisco and Telstra (which have agreed to a one-year lock-up according to KludeIn’s SEC filings). Company data from PitchBook notes that Near had tried but cancelled a fundraise in May 2021.

All in all, Near is an interesting example when considering the predicament that a lot of later-stage startups might be finding themselves at the moment.

On the one hand, the company has some big customers and some potentially interesting technology, especially in light of the swing from regulators and the public toward demanding more privacy in data intelligence products overall.

It works with major brands and companies including McDonald’s, Wendy’s, Ford, the CBRE Group and 60% of the Fortune 500, which use Near’s interactive, cloud-based AI platform (branded Allspark) to tap into anonymised, location-based profiles of users based on a trove of information that Near sources and then merges from phones, data partners, carriers and its customers. It claims the database has been built “with privacy by design.”

It describes its approach as “stitching” and says that it’s patent-protected, giving it a kind of moat against other competitors, and potentially some value as an asset for others that are building big data businesses and need more privacy-based approaches.

On the other hand, while financials detailed in KludeIn’s SEC filings show growth, it is at a very modest pace — numbers may not look that great to investors especially in the current climate. In 2020, Near posted revenues of $33 million, with estimated revenues of $46 million for 2021, $63 million for 2022 and $91 million for 2023. The company estimates that its gross profit margin for this year will be 72% ($44 million) but equally estimates that EBITDA has been negative and will continue to be until at least 2024.

Image Credits: Near

Looking out further than Near, it will be interesting to see how many others follow the company in taking the SPAC exit route, which has proven to be a controversial vehicle overall.

On the plus side, SPACs are lauded by supporters for being a faster, more efficient route for strong startups to enter the public markets and thus raise money from more investors (and giving sight of an exit to private investors): this is very much the position Near and KludeIn are taking.

“Enterprises around the world have trusted Near to answer their critical questions that help drive and grow their business for more than a decade. The market demand for data around human movement and consumer behavior to understand changing markets and consumers is growing exponentially and now is the time to accelerate the penetration of the large and untapped $23 billion TAM,” Anil Mathews, founder and CEO of Near, said in a statement. “Going public provides us the credibility and currency to double-down on growth and to continue executing on our winning flywheel for enhanced business outcomes over the next decade.”

“I am thrilled to partner with Anil and the entire team at Near as they continue to help global enterprises better understand consumer behavior and derive actionable intelligence from their global, full-stack data intelligence platform,” added Narayan Ramachandran, the chairman and CEO of KludeIn. “We believe this merger is highly compelling based on the diversified global customer base, superior SaaS flywheel and network effects of Near’s business, highlighted by the company’s strong customer net retention.”

On the minus side, those positives are also the very reasons for some of SPAC’s problems: Simply put, they have enabled public listings for companies that might have found it much harder, if not impossible, to do so through the scrutiny of more traditional channels. Sometimes that has played out okay anyway, but sometimes it has ended badly for everyone. Just this week, Enjoy — which also listed by way of a SPAC — said that it was on course to run out of money by June and was reviewing its strategic options.

Time, the appetite for more data intelligence and potentially some factors out of its control like the investment climate, ultimately will show which way Near will go. The transaction is expected to generate $268 million of gross proceeds, assuming there are no redemptions and a successful private placement of $95 million of KludeIn common stock, KludeIn said.

We (skim)read Meta’s metaverse manifesto so you don’t have to…

Meta’s recently crowned president of global affairs, Nick Clegg — who, in a former life, was literally the deputy prime minister of the U.K. — has been earning his keep in California by penning an approximately 8,000-word manifesto to promo “the metaverse”: aka, the sci-fi-inspired vapourware the company we all know as Facebook fixed on for a major rebranding last fall.

Back then, founder and CEO Mark Zuckerberg, pronounced that the new entity (Meta) would be a “metaverse-first” company “from now on”. So it’s kinda funny that the key question Clegg says he’s addressing in his essay is “what is the metaverse” — and, basically, why should anyone care? But trying to explain such core logic is apparently keeping Meta’s metamates plenty busy.

The Medium post Clegg published yesterday warns readers it will require 32 minutes of their lives to take in. So few people may have cared to read it. As a Brit, I can assure you, no one should feel obliged to submit to 32 minutes of Nick Clegg — especially not bloviating at his employer’s behest. So TechCrunch took that bullet for the team and read (ok, skim-read) the screed so you don’t have to.

What follows is our bullet-pointed digest of Clegg’s metaverse manifesto. But first we invite you to chew over this WordCloud (below), which condenses his ~7,900-word essay down to 50 — most boldly featuring the word “metaverse” orbiting “internet”, thereby grounding the essay firmly in our existing digital ecosystem.

Glad we could jettison a few thousand words to arrive at that first base. But, wait, there’s more!

Image credits: Natasha Lomas/TechCrunch

Fun found word pairs that leap out of the CleggCloud include “companies rules” (not democratic rules then Clegg?); “people technologies” (possibly just an oxymoron; but we’re open to the possibility that it’s a euphemistic catch-all for ill-fated startups like HBO’s Silicon Valley‘s (satirical) ‘Human Heater’); “around potential” (not actual potential then?); “meta physical” (we lol’d); and — squint or you’ll miss it! — “privacy possible” (or possibly “possible privacy”).

The extremely faint ink for that latter pairing adds a fitting layer of additional uncertainty that life in the Zuckerberg-Clegg metaverse will be anything other than truly horrific for privacy. (Keen eyed readers may feel obligated to point out that the CleggCloud also contains “private experience” as another exceptionally faint pairing. Albeit, having inhaled the full Clegg screed, we can confirm he’s envisaging “private experience” in exceptional, siloed, close-friend spaces — not that the entire metaverse will be a paradise for human privacy. Lol!)

Before we move on to the digest, we feel it’s also worth noting a couple of words that aren’t used in Clegg’s essay — and so can only be ‘invisibly inked’ on our wordcloud (much like a tracking pixel) — deserving a mention by merit of their omission: Namely, “tracking” and “profiling”; aka, how advertising giant Meta makes its money now. Because, we must assume, tracking and profiling is how Meta plans to make its money in the mixed reality future Clegg is trying to flog.

His essay doesn’t spare any words on how Meta plans to monetize its cash-burning ‘pivot’ or reconfigure the current “we sell ads” business model in the theoretical, mixed reality future scenario he’s sketching, where the digital commerce playground is comprised of a mesh of interconnecting services owned and operated by scores of different/competing companies.

But perhaps — and we’re speculating wildly here — Meta is envisaging being able to supplement selling surveillance-targeted ads by collecting display-rents from the cottage industry of “creators” Clegg & co. hope will spring up to serve these spaces by making digital items to sell users, such as virtual threads for their avatars, or virtual fitting rooms to buy real threads… (‘That’s a nice ‘Bored Ape T-Shirt’ you’re planning to sell — great job! — but if you want metamates to be able to see it in full glorious color you’ll want to pay our advanced display fees’, type thing. Just a thought!)

Now onwards to our digest of Clegg’s screed — which we’ve filleted into a series of bulleted assertions/suggestions being made by the Meta president (adding our commentary alongside in bold-italics). Enjoy how much time we’ve saved you.

  • There won’t be ‘a’ or ‘the metaverse’, in the sense of a single experience/owned entity; there will be “metaverse spaces” across different devices, which may — or may not — interoperate nicely [so it’s a giant rebranding exercise of existing techs like VR, AR, social gaming etc?] 
  • But the grand vision is “a universal, virtual layer that everyone can experience on top of today’s physical world” [aka total intermediation of human interaction and the complete destruction of privacy and intimacy in service of creating limitless, real-time commercial opportunities and enhanced data capture]
  • Metaverse spaces will over index on ephemerality, embodiment and immersion and be more likely to centre speech-based communication vs current social apps, which suggests users may act more candid and/or forget they’re not actually alone with their buddies [so Meta and any other mega corporates providing “metaverse spaces” can listen in to less guarded digital chatter and analyze avatar and/or actual body language to derive richer emotional profiles for selling stuff] 
  • The metaverse could be useful for education and training [despite the essay’s headline claim to answer “why it matters”, Clegg doesn’t actually make much of a case for the point of the metaverse or why anyone would actually want to fritter their time away in a heavily surveilled virtual shopping mall — but he includes some vague suggestions it’ll be useful for things like education or healthcare training. At one one point he enthuses that the metaverse will “make learning more active” — which implies he was hiding under a rock during pandemic school shutdowns. He also suggests metaverse tech will remove limits on learning related to geographical location — to which one might respond have you heard of books? Or the Internet? etc]
  • The metaverse will create new digital divides — given those who can afford the best hardware will get the most immersive experience [not a very equally distributed future then is it Clegg?] 
  • It’s anyone’s guess how much money the metaverse might generate — or how many jobs it could create! [🤷]
  • But! Staggeringly vast amounts of labor will be required to sustain these interconnected metaverse spaces [i.e. to maintain any kind of suspension of disbelief that it’s worth the time sink and to prevent them from being flooded with toxicity]
  • Developers especially there will be so much work for you!!! [developers, developers, developers!]
  • Unlike Facebook, there won’t be one set of rules for the metaverse — it’s going to be a patchwork of ToS [aka, it’ll be a confusing mess. Plus governments/states may also be doing some of the rule-making via regulation]
  • A lack of interoperability/playing nice between any commercial entities that build “metaverse experiences” could fatally fragment the seamless connectivity Meta is so keen on [seems inevitable tbh; thereby threatening the entire Meta rebranding project. Immersive walled gardens anyone?]
  • Meta’s metaverse might let you create temporary, siloed private spaces where you can talk with friends [but only in the same siloed way that FB Messenger offers E2EE via “Secret Conversations” — i.e. surveillance remains Meta’s overarching rule]
  • Bad metaverse experiences will probably be even more horrible than 2D-based cyberbullying etc [yep, virtual sexual assault is already a thing]
  • There are big challenges and uncertainties ahead for Meta [no shit]
  • It’s going to take at least 10-15 years for anything resembling Meta’s idea of connected metaverse/s to be built [Clegg actually specified: “if not longer”; imagine entire decades of Zuckerberg-Clegg!]
  • Meta hopes to work with all sorts of stakeholders as it develops metaverse technologies [aka, it needs massive buy-in if there’s to be a snowflake’s chance in hell of pulling off this rebranding pivot and not just sinking billions into a metaverse money-hole]
  • Meta names a few “priority areas” it says are guiding its metaverse development — topped by “economic opportunity” [just think of all those developer/creator jobs again! Just don’t forget who’s making the mega profits right now… All four listed priorities offer more PR soundbite than substance. For example, on “privacy” — another of Meta’s stated priorities — Clegg writes: “how we can build meaningful transparency and control into our products”. Which is a truly rhetorical ask from the former politician, since Facebook does not give users meaningful control over their privacy now — so we must assume Meta is planning a future of more of the same old abusive manipulations and dark patterns so it can extract as much of people’s data as it can get away with… Ditto “safety & integrity” and “equity & inclusion” under the current FB playbook.] 
  • “The metaverse is coming, one way or another” [Clegg’s concluding remark comes across as more of a threat than bold futuregazing. Either way, it certainly augurs Meta burning A LOT more money on this circus]

When big AI labs refuse to open source their models, the community steps in

Benchmarks are as important a measure of progress in AI as they are for the rest of the software industry. But when the benchmark results come from corporations, secrecy very often prevents the community from verifying them.

For example, OpenAI granted Microsoft, with which it has a commercial relationship, the exclusive licensing rights to its powerful GPT-3 language model. Other organizations say that the code they use to develop systems is dependent on impossible-to-release internal tooling and infrastructure or uses copyrighted data sets. While motivations can be ethical in nature — OpenAI initially declined to release GPT-2, GPT-3’s predecessor, out of concerns that it might be misused — by the effect is the same. Without the necessary code, it’s far harder for third-party researchers to verify an organization’s claims.

“This isn’t really a sufficient alternative to good industry open-source practices,” Columbia computer science Ph.D. candidate Gustaf Ahdritz told TechCrunch via email. Ahdritz is one of the lead developers of OpenFold, an open source version of DeepMind’s protein structure-predicting AlphaFold 2. “It’s difficult to do all of the science one might like to do with the code DeepMind did release.”

Some researchers go so far as to say that withholding a system’s code “undermines its scientific value.” In October 2020, a rebuttal published in the journal Nature took issue with a cancer-predicting system trained by Google Health, the branch of Google focused on health-related research. The coauthors noted that Google withheld key technical details including a description of how the system was developed, which could significantly impact its performance.

OpenFold

Image Credits: OpenFold

In lieu of change, some members of the AI community, like Ahdritz, have made it their mission to open source the systems themselves. Working from technical papers, these researchers painstakingly try to recreate the systems, either from scratch or building on the fragments of publicly available specifications.

OpenFold is one such effort. Begun shortly after DeepMind announced AlphaFold 2, the goal is to verify that AlphaFold 2 can be reproduced from scratch and make available components of the system that might be useful elsewhere, according to Ahdritz.

“We trust that DeepMind provided all the necessary details, but … we don’t have [concrete] proof of that, and so this effort is key to providing that trail and allowing others to build on it,” Ahdritz said. “Moreover, originally, certain AlphaFold components were under a non-commercial license. Our components and data — DeepMind still hasn’t published their full training data — are going to be completely open-source, enabling industry adoption.”

OpenFold isn’t the only project of its kind. Elsewhere, loosely-affiliated groups within the AI community are attempting implementations of OpenAI’s code-generating Codex and art-creating DALL-E, DeepMind’s chess-playing AlphaZero, and even AlphaStar, a DeepMind system designed to play the real-time strategy game StarCraft 2. Among the more successful are EleutherAI and AI startup Hugging Face’s BigScience, open research efforts that aim to deliver the code and datasets needed to run a model comparable (though not identical) to GPT-3.

Philip Wang, a prolific member of the AI community who maintains a number of open source implementations on GitHub, including one of OpenAI’s DALL-E, posits that that open-sourcing these systems reduces the need for researchers to duplicate their efforts.

“We read the latest AI studies, like any other researcher in the world. But instead of replicating the paper in a silo, we implement it open source,” Wang said. “We are in an interesting place at the intersection of information science and industry. I think open source is not one-sided and benefits everybody in the end. It also appeals to the broader vision of truly democratized AI not beholden to shareholders.”

Brian Lee and Andrew Jackson, two Google employees, worked together to create MiniGo, a replication of AlphaZero. While not affiliated with the official project, Lee and Jackson — being at Google, DeepMind’s initial parent company — had the advantage of access to certain proprietary resources.

MiniGo

Image Credits: MiniGo

“[Working backward from papers is] like navigating before we had GPS,” Lee, a research engineer at Google Brain, told TechCrunch via email. “The instructions talk about landmarks you ought to see, how long you ought to go in a certain direction, which fork to take at a critical juncture. There’s enough detail for the experienced navigator to find their way, but if you don’t know how to read a compass, you’ll be hopelessly lost. You won’t retrace the steps exactly, but you’ll end up in the same place.”

The developers behind these initiatives, Ahdritz and Jackson included, say that they’ll not only help to demonstrate whether the systems work as advertised but enable new applications and better hardware support. Systems from large labs and companies like DeepMind, OpenAI, Microsoft, Amazon, and Meta are typically trained on expensive, proprietary datacenter servers with far more compute power than the average workstation, adding to the hurdles of open-sourcing them.

“Training new variants of AlphaFold could lead to new applications beyond protein structure prediction, which is not possible with DeepMind’s original code release because it lacked the training code — for example, predicting how drugs bind proteins, how proteins move, and how proteins interact with other biomolecules,” Ahdritz  said. “There are dozens of high-impact applications that require training new variants of AlphaFold or integrating parts of AlphaFold into larger models, but the lack of training code prevents all of them.”

“These open-source efforts do a lot to disseminate the “working knowledge” about how these systems can behave in non-academic settings,” Jackson added. “The amount of compute needed to reproduce the original results [for AlphaZero] is pretty high. I don’t remember the number off the top of my head, but it involved running about a thousand GPUs for a week. We were in a pretty unique position to be able to help the community try these models with our early access to the Google Cloud Platform’s TPU product, which was not yet publicly available.”

Implementing proprietary systems in open source is fraught with challenges, especially when there’s little public information to go on. Ideally, the code is available in addition to the data set used to train the system and what are called weights, which are responsible for transforming data fed to the system into predictions. But this isn’t often the case.

For example, in developing OpenFold, Ahdritz and team had to gather information from the official materials and reconcile the differences between different sources, including the source code, supplemental code, and presentations that DeepMind researchers gave early on. Ambiguities in steps like data prep and training code led to false starts, while a lack of hardware resources necessitated design compromises.

“We only really get a handful of tries to get this right, lest this drag on indefinitely. These things have so many computationally intensive stages that a tiny bug y can greatly set us back, such that we had to retrain the model and also regenerate lots of training data,” Ahdritz said. “Some technical details that work very well for [DeepMind] don’t work as easily for us because we have different hardware … In addition, ambiguity about what details are critically important and which ones are selected without much thought makes it hard to optimize or tweak anything and locks us in to whatever (sometimes awkward) choices were made in the original system.”

So, do the labs behind the proprietary systems, like OpenAI, care that their work is being reverse-engineered and even used by startups to launch competing services? Evidently not. Ahdritz says the fact that DeepMind in particular releases so many details about its systems suggests it implicitly endorses the efforts, even if it hasn’t said so publicly.

“We haven’t received any clear indication that DeepMind disapproves or approves of this effort,” Ahdritz said. “But certainly, no one has tried to stop us.”

Fetcher raises $27M to automate aspects of job candidate sourcing

Reflecting the growing investor interest in HR technology startups, Fetcher, the talent acquisition platform formerly known as Scout, today closed a $27 million Series B funding round led by Tola Capital with participation from G20 Ventures, KFund, and Accomplice. The new money — $7 million in debt and $20 million in equity — brings the startup’s total capital raised to $40 million, which co-founder and CEO Andres Blank says is being put toward international expansion and building out the Fetcher platform with new applicant tracking system (ATS) integrations and customer relationship management capabilities.

Fetcher was co-launched in 2014 by Blank, Chris Calmeyn, Javier Castiarena, and Santi Aimetta as a professional networking app called Caliber. After a few years, the founding Fetcher team decided to pivot into recruitment, leveraging some of the automation technology they’d built into Caliber.

“Hiring high-quality, diverse candidates had always been a pain point for me. At one of my prior startups, I personally experienced this issue, and after bringing on a recruiting team to help scale hiring efforts, I saw that their time was also too valuable to be spent on the manual, repetitive tasks that come with sourcing candidates,” Blank told TechCrunch in an email interview. “Rather than relying on expensive staffing fees, I thought there must be a better way to keep sourcing in-house, without it taking up too much time and energy on the talent acquisition teams and hiring managers.”

Through a Chrome extension, Fetcher’s platform ties in with ATS products as well as Gmail and Outlook to allow recruiters to source candidates directly from LinkedIn. Fetcher filters jobseekers into prebuilt email workflows, offering analytics including progress toward diversity goals at the individual, team, position, and company levels.

Fetcher

The Fetcher candidate directory.

Fetcher also performs predictive modeling, automatically gauging the interest of job candidates from their replies, and “automated sourcing,” which runs in the background to push applicants through vetting processes via automated emails.

“A great candidate experience is essential for any company, and part of that experience comes from building long-term relationships with candidates over time. Fetcher’s candidate directory allows companies to remarket to qualified candidates, set up reminders for future connections, and add additional outreach emails to the automated sequences,” Blank said. “Overall, the goal is to make it simple for companies to store, update, and connect with great candidates over time, messaging them about future job opportunities, milestones at the company, and more.”

The reliance on algorithms is a bit concerning, given the potential for bias — Amazon infamously scrapped a recruitment algorithm that favored male engineers and New York City recently placed restrictions on the use of AI in hiring. When asked about it, Blank asserted that the platform’s automation technologies allow for “a more diverse group of prospects” to push through the hiring funnel. He also highlighted Fetcher’s outreach policy, noting that people who don’t wish to be contacted about opportunities via Fetcher can send data deletion requests.

“[O]ur secret sauce here at Fetcher is combining both machine and human intelligence in order to minimize the biases that exist on both sides,” Blank said. “Beyond this, we also have diversity metrics on each search (visible on our platform to the client too), which keeps us in check. If we’re over- or under-indexing anywhere on the gender or demographics front, the platform can course correct. Finally, we remove selection biases from the client. The way we do this is that once a client trusts that the search is heading in the right direction (after vetting a handful of candidates upfront), they place the search on full automation. This means that going forward, they are no longer vetting every candidate, but simply reaching out to all qualified candidates that are found for [a given] open role.”

Blank linked to case studies from customers like Frame.io, which recently used Fetcher to hire employees mostly from underrepresented groups. But biases can enter at many different, often unpredictable stages of the pipeline. As Harvard Business Review’s Miranda Bogen writes: “For example, if [a] system notices that recruiters happen to interact more frequently with white men, it may well find proxies for those characteristics (like being named Jared or playing high school lacrosse) and replicate that pattern. This sort of adverse impact can happen without explicit instruction, and worse, without anyone realizing.”

Fetcher

Image Credits: Fetcher

The risk doesn’t appear to be dissuading recruiters. Fetcher currently has over 350 customers (growing 10% month-over-month) including Behr Paint, Albertson’s, Foursquare, and Shutterstock., and annual recurring revenue tripled in the last 12 months.

Beyond the strong top-line numbers, Fetcher is benefiting from the broader boom in the HR tech segment, which has seen high venture capital activity over the past few months. According to Pitchbook, HR tech startups collected more than $9.2 billion in venture capital funding globally from January 2021 to October 2021 — a 130% jump from 2020’s total.

“Fetcher is uniquely positioned as one of the only software-as-a-service recruiting platforms to automate both candidate sourcing and email outreach efficiently,” Blank said. “Rather than using a straight database model, Fetcher is the only sourcing solution that can truly automate the sourcing process for companies, based on its unique combination of ‘machine learning with human intelligence.’ This model allows for what feels like a 24/7 sourcer to work in the background for each client. By automating both the sourcing and outreach sides of recruiting, Fetcher can reduce the number of internal sourcers and recruiters a company needs, as well as significantly reduce the budget being spent on outside recruiting firms, agencies, or consultants.”

Fetcher employs 45 people, currently, and plans to double that number by the end of the year.

Everstream Analytics secures new cash to predict supply chain disruptions

Everstream Analytics, a supply chain insights and risk analytics startup, today announced that it raised $24 million in a Series A round led by Morgan Stanley Investment Management with participation from Columbia Capital, StepStone Group, and DHL. CEO Julie Gerdeman said that the new money would be used to “propel technology innovation” and “further global expansion.”

Everstream, which was launched as Resilience360 and Riskpulse, provides predictive insights for supply chains. Drawing on billions of supply chain interactions, the company applies AI to assess materials, suppliers, and facilities for risk.

Plenty of startups claim to do this, including Backbone, Altana, and Craft. Project44 recently raised $202 million to expand its own set of predictive analytics tools, including estimated time of arrivals for shipments.

But what sets Everstream apart is its access to proprietary data that goes beyond what competitors are leveraging, according to Gerdeman.

“[Everstream provides] visibility into essentially every network, component, ingredient, ​and raw material around the world,” she told TechCrunch via email. “Connected business networks, scalable computing power, graph data base technology, and advances in AI algorithms enable Everstream to combine massive volumes of public and proprietary data to build a model of the global supply chain.”

As new data enters the platform, Everstream, which integrates with existing enterprise resource planning systems, retrains its AI system to reflect the current supply chain environment. Customers receive proactive warnings based on signals including financial reports and news of weather events, environmental and sustainability risks, and natural disasters.

For example, Everstream can warn businesses when it might be difficult to source a specific material and how likely customers are to cancel, increase, or move forward orders. It can also provide suggestions for optimizing logistics operations based on metrics such as timeliness, quality, and cost of goods shipped.

“Everstream’s AI-based models and preset dynamic thresholds can be used to predict disruptions and prescribe recommendations to mitigate risk and deliver better results to the business needs,” Gerdeman added. “[Everstream] identifies the most impactful risks in the network and creates targeted insights-based on inputs from the … platform, including incident monitoring, predictive risks, ESG, and shipment data — slashing time, cost, and complexity.”

Most would argue these are useful tools at a time when uncertainty continues to dog the supply chain — assuming Everstream’s AI systems perform as well as advertised. While some surveys show tepid adoption of predictive analytics among the supply chain industry, Gartner recently found that 87% of supply chain professionals plan to invest in “resilience” within the next two years, including automation and AI.

Investors seemingly see the potential. Last year was a banner year for venture-backed supply chain management companies, which saw $11.3 billion in funding, according to Crunchbase.

For its part, Everstream claims its customer base has grown 550% to date in 2022 and now includes brands like AB InBev, Google, Bayer, Schneider Electric, Unilever, and Whirlpool. Mum’s the word on concrete revenue numbers; Gerdeman demurred when asked about them.

“The pandemic has illustrated why deep visibility is needed not only into a company’s network, but down to the component, ingredient, ​and raw material level, because it doesn’t matter if the company’s supplier is operational if their suppliers are not,” Gerdeman said. “Everstream’s insights are not only predictive in nature, but they are also prescriptive – meaning we not only tell clients what’s coming next, but also what they should do about it.”

Everstream, which employs 100 people, has raised $70 million in equity and debt funding so far.

Behold NeuroMechFly, the best fruit fly simulator yet

Drosophila melanogaster, the common fruit fly, is in some ways a simple creature. But in others it is so complex that, as with any form of life, we are only scratching the surface of understanding it. Researchers have taken a major step with D. melanogaster by creating the most accurate digital twin yet — at least in how it moves and, to a certain extent, why.

NeuroMechFly, as the researchers at EPFL call their new model, is a “morphologically realistic biomechanical model” based on careful scans and close observation of actual flies. The result is a 3D model and movement system that, when prompted, does things like walk around or respond to certain basic stimuli pretty much as a real fly would.

To be clear, this isn’t a complete cell-by-cell simulation, which we’ve seen some progress on in the last few years with much smaller microorganisms. It doesn’t simulate hunger, or vision or any sophisticated behaviors — not even how it flies, only how it walks along a surface and grooms itself.

What’s so hard about that, you ask? Well, it’s one thing to approximate this type of movement or behavior and make a little 3D fly that moves more or less like a real one. It’s another to do so to a precise degree in a physically simulated environment, including a biologically accurate exoskeleton, muscles, and a neural network analogous to the fly’s that controls them.

To make this very precise model, they started with a CT scan of a fly, in order to create the morphologically realistic 3D mesh. Then they recorded a fly walking in very carefully controlled circumstances and tracked its precise leg movements. EPFL researchers then needed to model exactly how those movements corresponded to the physically simulated “articulating body parts, such as head, legs, wings, abdominal segments, proboscis, antennae, halteres,” the latter of which is a sort of motion-sensing organ that helps during flight.

Image Credits: Pavan Ramdya (EPFL)

They showed that these worked by bringing the precise motions of the observed fly into a simulation environment and replaying them with the simulated fly — the real movements mapped correctly onto the model’s. Then they demonstrated that they could create new gaits and movements based on these, letting the fly run faster or in a more stable way than what they had observed.

Image Credits: Pavan Ramdya (EPFL)

Not that they’re improving on nature, exactly; they’re just showing that the simulation of the fly’s movement extended to other, more extreme examples. Their model was even robust against virtual projectiles…to a certain extent, as you can see in the animation above.

“These case studies built our confidence in the model. But we are most interested in when the simulation fails to replicate animal behavior, pointing out ways to improve the model,” said EPFL’s Pavan Ramdya, who leads the group that built the simulator (and other D. melanogaster–related models). Seeing where their simulation breaks down shows where there’s work to do.

“NeuroMechFly can increase our understanding of how behaviors emerge from interactions between complex neuromechanical systems and their physical surroundings,” reads the abstract of the paper published last week in Nature Methods. By better understanding how and why a fly moves the way it does, we can understand the systems that underlie it better as well, producing insights in other areas (fruit flies are among the most used experimental animals). And of course if we ever wanted to create an artificial fly for some reason, we would definitely want to know how it works first.

While NeuroMechFly is in some ways a huge advance in the field of digitally simulating life, it’s still (as its creators would be the first to acknowledge) incredibly limited, focusing solely on specific physical processes and not on the many other aspects of the tiny body and mind that make a Drosophila a Drosophila. You can check out the code and perhaps contribute over at GitHub or Code Ocean.