Connect with us

Tech

Why Cohere’s ex-AI research lead is betting against the scaling race

Published

on


AI labs are racing to build data centers as large as Manhattan, each costing billions of dollars and consuming as much energy as a small city. The effort is driven by a deep belief in “scaling” — the idea that adding more computing power to existing AI training methods will eventually yield superintelligent systems capable of performing all kinds of tasks.

But a growing chorus of AI researchers say the scaling of large language models may be reaching its limits, and that other breakthroughs may be needed to improve AI performance.

That’s the bet Sara Hooker, Cohere’s former VP of AI Research and a Google Brain alumna, is taking with her new startup, Adaption Labs. She co-founded the company with fellow Cohere and Google veteran Sudip Roy, and it’s built on the idea that scaling LLMs has become an inefficient way to squeeze more performance out of AI models. Hooker, who left Cohere in August, quietly announced the startup this month to start recruiting more broadly.

In an interview with TechCrunch, Hooker says Adaption Labs is building AI systems that can continuously adapt and learn from their real-world experiences, and do so extremely efficiently. She declined to share details about the methods behind this approach or whether the company relies on LLMs or another architecture.

“There is a turning point now where it’s very clear that the formula of just scaling these models — scaling-pilled approaches, which are attractive but extremely boring — hasn’t produced intelligence that is able to navigate or interact with the world,” said Hooker.

Adapting is the “heart of learning,” according to Hooker. For example, stub your toe when you walk past your dining room table, and you’ll learn to step more carefully around it next time. AI labs have tried to capture this idea through reinforcement learning (RL), which allows AI models to learn from their mistakes in controlled settings. However, today’s RL methods don’t help AI models in production — meaning systems already being used by customers — learn from their mistakes in real time. They just keep stubbing their toe.

Some AI labs offer consulting services to help enterprises fine-tune their AI models to their custom needs, but it comes at a price. OpenAI reportedly requires customers to spend upwards of $10 million with the company to offer its consulting services on fine-tuning.

Techcrunch event

San Francisco
|
October 27-29, 2025

“We have a handful of frontier labs that determine this set of AI models that are served the same way to everyone, and they’re very expensive to adapt,” said Hooker. “And actually, I think that doesn’t need to be true anymore, and AI systems can very efficiently learn from an environment. Proving that will completely change the dynamics of who gets to control and shape AI, and really, who these models serve at the end of the day.”

Adaption Labs is the latest sign that the industry’s faith in scaling LLMs is wavering. A recent paper from MIT researchers found that the world’s largest AI models may soon show diminishing returns. The vibes in San Francisco seem to be shifting, too. The AI world’s favorite podcaster, Dwarkesh Patel, recently hosted some unusually skeptical conversations with famous AI researchers.

Richard Sutton, a Turing award winner regarded as “the father of RL,” told Patel in September that LLMs can’t truly scale because they don’t learn from real world experience. This month, early OpenAI employee Andrej Karpathy told Patel he had reservations about the longterm potential of RL to improve AI models.

These types of fears aren’t unprecedented. In late 2024, some AI researchers raised concerns that scaling AI models through pretraining — in which AI models learn patterns from heaps of datasets — was hitting diminishing returns. Until then, pretraining had been the secret sauce for OpenAI and Google to improve their models.

Those pretraining scaling concerns are now showing up in the data, but the AI industry has found other ways to improve models. In 2025, breakthroughs around AI reasoning models, which take additional time and computational resources to work through problems before answering, have pushed the capabilities of AI models even further.

AI labs seem convinced that scaling up RL and AI reasoning models are the new frontier. OpenAI researchers previously told TechCrunch that they developed their first AI reasoning model, o1, because they thought it would scale up well. Meta and Periodic Labs researchers recently released a paper exploring how RL could scale performance further — a study that reportedly cost more than $4 million, underscoring how expensive current approaches remain.

Adaption Labs, by contrast, aims to find the next breakthrough, and prove that learning from experience can be far cheaper. The startup was in talks to raise a $20 million to $40 million seed round earlier this fall, according to three investors who reviewed its pitch decks. They say the round has since closed, though the final amount is unclear. Hooker declined to comment.

“We’re set up to be very ambitious,” said Hooker, when asked about her investors.

Hooker previously led Cohere Labs, where she trained small AI models for enterprise use cases. Compact AI systems now routinely outperform their larger counterparts on coding, math, and reasoning benchmarks — a trend Hooker wants to continue pushing on.

She also built a reputation for broadening access to AI research globally, hiring research talent from underrepresented regions such as Africa. While Adaption Labs will open a San Francisco office soon, Hooker says she plans to hire worldwide.

If Hooker and Adaption Labs are right about the limitations of scaling, the implications could be huge. Billions have already been invested in scaling LLMs, with the assumption that bigger models will lead to general intelligence. But it’s possible that true adaptive learning could prove not only more powerful — but far more efficient.

Marina Temkin contributed reporting.



Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Tech

Waymo starts autonomous testing in Philadelphia

Published

on

By


Waymo is adding another four cities to its growing list of robotaxi rollouts. The company announced Wednesday it has begun testing its autonomous vehicles (with a safety monitor) in Philadelphia, and that it will start manual driving to collect data in Baltimore, St. Louis, and Pittsburgh.

Waymo did not offer a timeline for when it plans to launch commercial services in those locations, nor do we know whether the Alphabet-owned company will partner with other companies to operate robotaxis in each one. That has been the move in cities like Atlanta and Austin, for example, where Waymo has partnered with Uber to advance its robotaxi rollout.

But the new locations join a list of over 20 cities where the company is either offering rides, prepping a commercial launch, or testing. Waymo is also now offering rides on freeways in Los Angeles, Phoenix, and the San Francisco Bay Area. The company plans to be doing one million rides per week by the end of 2026.

Waymo has done all this while claiming to be operating at a level five times safer than humans, according to data the company recently released.

But the expansion has not come without its issues. The National Highway Traffic Safety Administration is investigating how the company’s vehicles operate near school buses, after a Waymo was filmed driving around a stopped bus in Atlanta in September.

This week, Austin news outlet KXAN published a report showing Waymo’s vehicles have driven past school buses that were in the process of unloading or loading children multiple times — including after Waymo claims to have shipped software updates to address the problem.

Techcrunch event

San Francisco
|
October 13-15, 2026

Continue Reading

Tech

Spotify Wrapped 2025 adds its first multiplayer feature with ‘Wrapped Party’

Published

on

By


Spotify Wrapped is back. After last year’s widely criticized flop that included an AI podcast as its highlight, the streamer’s highly anticipated annual review feature has returned to its roots. This year, Spotify is doubling down on what it knows works best: deep dives into your streaming data, creative experiences, messages from favorite artists, and other social features.

The company claims that Wrapped 2025 is its biggest, as it’s introducing nearly a dozen new features in addition to its old standbys, like top songs and artists. Plus, it’s offering more visibility into users’ data than in years past. For the first time, Spotify Wrapped is adding a live multiplayer feature to compare your listening data with friends.

Wrapped Party, Wrapped’s first live interactive experience, allows you to invite up to nine friends to compare listening stats.

Image Credits:Spotify

Also new this year, your Top Songs Playlist will include the play counts for each of the top songs, so you can actually see how much time you spent with your favorite tracks.

Other standout features this year include an interactive Top Song Quiz, a Listening Age feature, and Wrapped Clubs, which match you to one of six unique listening styles.

The company believes these additions will not only bring back the personalized, engaging experience that users have long expected from Wrapped, but will take it a step further by making it more interactive than before.

In the Top Song Quiz, for instance, you can try to guess which top song soundtracked your year before seeing the results.

Techcrunch event

San Francisco
|
October 13-15, 2026

Image Credits:Spotify

The new interactive Wrapped Party feature isn’t just about comparing the personal streaming data you’ve already received to your friends’ data, as that’s something people already do on social media. Instead, the feature presents unique data stories for your group, like who’s the “most obsessed fan,” the “early bird,” the most “picky listener,” or even something as nice as the “dinner table explainer,” meaning the person who listens to the most news podcasts.

Image Credits:Spotify

Spotify says these awards update dynamically every time you join a Wrapped Party, so no two sessions are ever the same — even if you run through them again with the same group of friends.

The new Wrapped Clubs, meanwhile, will group you into one of half a dozen listening styles, like the “Soft Hearts Club,” the “Club Serotonin,” the “Full Charge Crew,” the “Cosmic Stereo Club,” and others. You’ll also receive a role in the club based on your listening data. You might be a club leader if your listening choices strongly matches the club’s values, a scout if you’re always seeking out new releases, or an archivist if you listen to music from past eras.

Image Credits:Spotify

Another feature, Listening Age, compares your 2025 music listening to others in your age group. To calculate your age, the feature considers the release years of the tracks you listen to most. From there, it identifies the five-year span of music that you engaged with more than other listeners your age.

Image Credits:Spotify

As in prior years, you’ll see your top songs, top artists, top genres, and, for the first time, top albums. If you engaged with audiobooks and podcasts, you’ll see metrics for those as well. Artists, writers, and podcasters will have their own version of Wrapped as before. And top fans will again receive video messages from their favorite artists, podcasters, and, now, authors.

You’ll also receive a playlist of your top songs of the year, as before.

Image Credits:Spotify

What you won’t find in this year’s Wrapped is any feature that advertises it was made with AI.

In a press briefing on Tuesday, Spotify’s Senior Director of Global Marketing, Matt Luhks, admitted the company received a “lot of feedback” about its 2024 AI-focused Wrapped experience, saying it was a “mix of positive and ‘more constructive feedback,’” despite the feature driving more engagement than prior years.

“We take all of that in. We use that as information, insights, [and] inspiration for how we approached Wrapped this year,” he said in a press event ahead of today’s launch.

“What our users tell us about Wrapped means a lot to us, so it was really informative in how we approached Wrapped this year. And what we tried to build was the most creative, most innovative, most engaging Wrapped ever,” he added, setting a high bar for the 2025 edition of the now 11-year-old annual year-in-review feature.

“We’re the original and, we believe, still the best,” Luhks said.

Image Credits:Spotify

Still, AI was a part of the Wrapped experience. Though the company claims the overall experience was not made with AI, it does leverage a LLM (large language model) to add a storytelling layer to Wrapped’s facts and figures, and natural language summaries in other parts of its experience, looking back on your data.

Spotify’s attempt to fix Wrapped after a notable stumble comes as the streamer faces increased competition from Apple, Amazon, YouTube, and others, which have all launched their own annual review features, inspired by Wrapped.

“Everyone seems to have their own version of Wrapped. Now, there’s a lot of reviews and replays and rewinds out there, but we believe that Wrapped still sets the bar for these year-end recaps,” Luhks said.

Along with the consumer experience, Spotify shared its top artists, songs, albums, podcasts, and audiobooks for the year, with top winners that included, respectively, Bad Bunny (top song and album), Joe Rogan (“The Joe Rogan Experience” podcast), and Rebeca Yarros (author of “Fourth Wing”).

Continue Reading

Tech

Nothing looks to its community to raise $5M, wants to be ‘IPO-ready’ in 3 years

Published

on

By


Hardware maker Nothing is letting its user base buy its stock as part of a new community investment round of $5 million. The new round, which opens on December 10, will enable consumers to buy the company’s shares at its Series C valuation of $1.3 billion.

The company said it has so far raised $8 million in total from over 8,000 people across two previous community investment rounds. It held its first community funding event in 2021, aiming to raise $1.5 million.

“This isn’t about raising capital, it’s about giving our community/fans a chance to invest while we’re private and join us on the journey,” a spokesperson for Nothing told TechCrunch.

Community investors have a rotating seat on the company’s board, but it is unclear what else they get for investing in the company through such rounds.

Nothing raised $200 million in its Series C back in September from investors including Tiger Global, GV, Highland Europe, EQT, Latitude, I2BF and Tapestry. The company has raised $450 million to date.

The community round comes as Nothing makes changes to its corporate structure as it tries to increase its share of a smartphone market dominated by giants like Samsung and Apple. The company is spinning off its budget CMF brand, and plans to explore AI-centric devices while it keeps building smartphones and audio products. And Nothing claims it crossed $1 billion in cumulative revenue this year, up 150% from 2024.

The startup is working to be “IPO-ready” in three years, CEO Carl Pei told TechCrunch in an email. “The timing will depend on market conditions and what makes sense for the business at that point in time,” he said.

Techcrunch event

San Francisco
|
October 13-15, 2026

“What’s important is that we’re already operating with that discipline now. We’re building the systems, the governance, the financial discipline that a public company needs. It forces us to think longer-term and make smarter decisions that prioritise sustainable growth,” Pei added.

It’s not clear if Nothing aims to raise another round before an IPO. When asked about its fundraising plans, a Nothing spokesperson said the company is not thinking about raising capital immediately, but it wouldn’t be averse to those conversations.

Those interested in investing in the community round can use platforms like Wefunder and Crowdcube to participate.

Continue Reading

Trending