The Gradient

Important announcement! Today we launch a new Twitter-like space for the AI community - Sigmoid Social. We hope to ensure the thriving AI Twitter community can live on by maintaining this Mastodon instance going forward. Join here: sigmoid.social

My essay 'After Orthogonality,' the culmination of 4 years of work on virtue-ethics, rationality, AI, decision-theory, and praxis is out on @gradientpub

New essay from @peligrietzer: What follows for AI alignment if we take the concept of eudaimonia — active, rational human flourishing — seriously?

@IasonGabriel is an inspiration to me -- he works across disciplines to arrive at deeper and more grounded views of 'AI alignment' and the ethics of AI more broadly -- both of which are often held in very confused/naive ways by the ML community. Bonus: he has a very pleasant voice -- looking forward to listening to this discussion with the talented @spaniel_bashir

🎙️Excited to share this new podcast with @spaniel_bashir at @gradientpub: open.substack.com/pub/thegradien… It’s one of the deepest conversations I’ve had about AI & philosophy and we got it all on tape! 📼

New @gradientpub interview out (after a while!) with @IasonGabriel on value alignment, theories of justice, questions in political philosophy and their influence on Iason's thinking!

What do disembodied notions of AGI miss? @superspeeg writes: “Instead of trying to glue modalities together into a patchwork AGI, we should pursue approaches to intelligence that treat embodiment and interaction with the environment as primary, and see modality-centered processing as emergent phenomena.”

AGI won’t emerge from piling on modalities. It must be embodied, interactive, and unified –– where perception and action are core, and “modality” is just a side-effect. We need to stop building patchwork solutions and start building minds. New piece in @gradientpub: AGI Is Not Multimodal. Some bits 🧵👇

THE @AIRSTREET 2024 YEAR IN REVIEW We began last year’s review with a look back of the preceding 12 months: 2023 was a year of both immense technological acceleration and constant flux. AI truly exploded into the mainstream: industries witnessed the birth of potentially existential competition overnight and never before seen volumes of private and corporate capital flowed into large-scale AI endeavors. Esoteric topics such as OpenAI’s corporate structure shakeup achieved mindshare parity with the colossal implosions of FTX and Silicon Valley Bank. All of this happened against a backdrop of tightening private markets, a 20 year record high ending for the Nasdaq, and shifting political agendas. And yet, as we continue to repeat, it is still arguably day 1 for AI. As we now review the 12 months of 2024, it’s hard to say that much has changed! Billions of dollars have flowed into AI companies. We had another debate about OpenAI’s corporate structure. Public technology markets ripped again. The world continues to figure out the downstream effects of general purpose AI technology, and significant vibe shifts unfolded. The one update we’d make is on crypto - we’ve gone from the dark days of the FTX collapse to Bitcoin hit an all-time high, reaching beyond $100k. Yet another vibe shift… For Air Street, 2024 has been another great year of: new investments, a new installment of the State of AI Report, and record-breaking output and engagement with our content and community work. In the spirit of building in the open and holding ourselves accountable to our progress, here is our year in review for 2024. We cover Air Street, Air Street Press, the State of AI Report, and our community engagement. **PORTFOLIO UPDATES** In 2024, we made five new investments and one follow-on investment in Fund 2, several of which have not been announced, which brings its portfolio to ten active companies. We exited three companies from Fund 1 (Exscientia, Adept, and Graphcore) by way of M&A. Several additional rounds have not yet been announced. Now turning to the portfolio highlights: 🔬 @cellvoyant, Jan ‘24 £7.6M Seed. CellVoyant is developing AI-first live cell imaging products to predict and optimize the differentiation of stem cells, which give rise to all mature cells in the human body. By doing so, they’re making it possible to controllably manufacture any cell and tissue in the body, at scale. @rafacarazosalas 👷 @intenseye, Feb ‘24 $64M Series B. The company accelerated its market leadership in applying computer vision to safeguard employee health and safety by detecting 65 million unsafe acts and conditions in 2024 alone. Operating in 25+ countries across 6 continents, Intenseye protects over 100,000 workers every day in facilities at Amazon, Coca Cola, and Siemens from dangerous situations that otherwise hide in plain sight. @sercanesn 👾 @Interloom, Mar ‘24 $3M day 1 funding round. Led by the team that built enterprise knowledge graph software, Boxplot, which was acquired by Hyperscience, Interloom is helping businesses adopt AI-driven process automation. The company’s automation software is designed to maximize the value of human and machine task orchestration. Stay tuned this year for more unveiling! @fabianjakobi 🧬 @ProfluentBio, Mar ‘24 $35M funding round. Profluent is the premier organization for frontier AI models in biology. A few weeks after their $35M raise, the team unveiled OpenCRISPR-1, the world’s first open source AI-generated gene editor that is nearly 200 mutations away from any other known natural CRISPR-associated protein. This work, profiled by the New York Times, has attracted engagement from all major biopharma giants while researchers are making use of the editor in labs today. @thisismadani 🔬 @labgeni_us, May ‘24 £35M Series B. LabGenius is advancing its wholly-owned pipeline of multispecific therapeutic antibodies to address the challenge of on-target, off-tumour toxicity. The company completed a successful collaboration with Sanofi and will expand the scope of its automated AI-first discovery platform. @jejfield 👾 @AdeptAILabs founders and team joins @amazon, Jun ‘24 Founders and several team members of Adept joined Amazon’s AGI organization to further their mission of building useful general intelligence. Later in Dec ‘24, Amazon announced that @jluan (Adept’s Co-Founder/CEO) and Pieter Abbeel (famed Berkeley Professor and Co-Founder of robotics startup, Covariant, whose founders Amazon has also recruited) are to lead a new Amazon AGI SF Lab focused on foundational technology to power useful AI agents. 🎥 @odysseyml, Jul ‘24 $8M inaugural round to build generative worlds for film and games. Following Odyssey’s $18M Series A in November, the team unveiled Explorer - a preview of their image-to-world model that transforms any image into an editable, detailed 3D generative world. In a big vote of confidence, the legendary Pixar Co-Founder, Ed Catmull, has joined the company’s board and invested. @olivercameron @jeffrey_hawke 🎮 Graphcore joins SoftBank, Jul ‘24 The company joined SoftBank as a fully-owned subsidiary to accelerate the group’s efforts in AI-focused semiconductors alongside its major investment in ARM. 🧬 @exscientiaAI joins @RecursionPharma, Aug ‘24 Exscientia, the NASDAQ-listed AI-first precision medicine company that we became shareholders in through their acquisition of Allcyte in 2021, was itself acquired by NASDAQ-listed AI-first drug discovery leader Recursion for $688M in a major power-up for AI-first biology. The deal completed in Nov ‘24 and results in a full-stack discovery and design company across biology and chemistry with the largest GPU compute cluster in biopharma. The business has the potential to read out 10 clinical trials across rare disease, precision oncology, and infectious diseases in the next 18 months. @RecursionChris 👨‍💻 @patinasystems, Sep ‘24 A day one funding round. The company is building tools to enable the creation of spontaneous software that could be customized and shared to suit your specific needs. You will no longer need to wait for your photos, camera, messaging and maps applications to receive updates suited to the average user. You’ll be able to build what you need with ease. @tylerangert 🤖 @poolsideai , Oct ‘24 $500M Series B. Poolside is building the most capable AI-first software development system as the first major economically impactful capability in which AI will reach and surpass human level capabilities. Their products are already being pressure-tested in the toughest environments and most demanding large enterprises. @eisokant @jasoncwarner We’re looking forward to sharing more detail on several additional undisclosed investments soon. Angel portfolio: Finally, a handful of my portfolio companies raised great rounds in 2024, including: 🤖 @ContextualAI - $80M Series A to build out their generative AI platform optimized end-to-end for enterprise retrieval augmented generation. @douwekiela 🏭 @CrusoeAI - $600M Series D at a $2.8B valuation to support the buildout of Crusoe’s vertically integrated AI cloud, powered by environmentally aligned energy sources. @ChaseLochmiller 🌱 @lifeschemistry - $130M Series C to advance drug compounds discovered in nature into clinical trials. The company cleared its first investigational new drug application for a first-in-class medicine to fight inflammation associated with atopic dermatitis. @viswacolluru 🛰️ @MuonSpace - $56M Series B, surpassing $100M in customer contracts in 2024 for end-to-end space systems. 📞 @polyaivoice - $50M Series C at a $500m valuation to accelerate its leadership in voice-based customer support. The company serves close to 100 enterprise customers, including FedEx, PG&E, Caesars, Marriott and Unicredit. @nikola_mrksic 🏎️ @wayve_ai - $1.05B Series C, one of the largest rounds of its kind, to scale its embodied AI for autonomous driving products. @alexgkendall **AIR STREET PRESS** Alongside our global events and the @stateofai, we publish in-depth essays and shorter-form content every week, as well as our monthly @guidetoai newsletter. This covers everything from trends in research, past and present case studies around the development and adoption of technology, as well as critical assessments into contested areas of policy. Transparency is important for our ecosystem. We therefore share the submissions we make to government bodies, as well as the notes we prepare ahead of any meetings with policymakers. We’ve continued to be vocal on the issue of defense by emphasising both investors’ responsibility to act and the importance of governments reforming procurement. The latter was the cornerstone of our submission to the UK Government’s Strategic Defence Review. This was also the theme of a panel @chalmermagne hosted at our friend @ericsles' European Defense Tech Summit, featuring Jonathan Turner from Accel and Lorenz Meier of Auterion. In February of this year, we brought all of this work together as @airstreetpress. We’d love to have you join nine thousand readers to receive all of this content in your inbox. Since then, we’ve written 30 analysis pieces, 7 policy essays, shared 9 pieces of major portfolio news and investments, 10 monthly issues of Guide to AI, and 6 posts about the State of AI and deeper dives into its outtakes. We saw close to 200,000 users hit the press and total views up to 346,000 during the year. Air Street Press is read from 49 US states and 122 countries, with a third of readers coming from the US, 18% from the UK, and 5% from each of Germany, India and France. We know this is a lot to read, so we’ve compiled our 2024 Greatest Hits album :-) A few personal highlights included: - The State of State of AI Report looking back on 5 years of AI progress - Essay on foundation model economics - Deep-dive into open-endedness and agentic AI - Examination of biorisk, biodefense, and associated policy issues. - My interview with @eisokant, the CTO and Co-Founder of Poolside on AI scaling laws. **STATE OF AI REPORT 2024** In October, we released the 7th edition of the State of AI Report, which covers the most important developments across research, industry, politics, and safety. We saw over 325,000 page views and close to 200,000 users visiting the site in the last year, with a third coming from the US, 9% from each of the UK and Germany, followed by Japan, India and France. To mark publication day, we held our second State of AI launch event in San Francisco, bringing 150 together founders, researchers, and operators for a night of discussion and best practice sharing. We covered large-scale model pre-training, complex reasoning, multi-modality, evaluations, inference optimization techniques, inference-time compute, truly open source AI (here’s a Q&A we did with Percy afterwards), and design best practices for AI-first product experiences. The report was dubbed ‘magisterial’ by Fortune: In the months that ensued, we enjoyed diving into the report’s findings on a few podcasts. Nathan recorded an edition of the MAD Podcast with @mattturck in person in NY. Nathan was also interviewed by Alex Heath on The Verge’s Decoder podcast about AI adoption by enterprise and why investors continue to subsidize frontier labs’ huge compute bills. For the third time, Nathan closed out the year with @spaniel_bashir on @gradientpub podcast. We spent two hours looking back on the year in AI, discussing the State of AI Report, our early impressions of o3, AGI, adoption, China, and a range of other topics. **COMMUNITY** 2024 saw our most active program of global AI events yet. We focus these events on a) best practices that entrepreneurs and researchers can learn from and b) attracting the very best contributors and heavily curating event attendance to make sure that everyone makes high-value connections. We share the list of attendees ahead of time so you’re able to get a head start on meeting those you want to connect with, and you can join our 350+ strong Air Street Community on Slack afterwards to keep in touch with folks. We started the year with our first installment of Munich AI, which brought together 200 guests, including speakers from Interloom, Sereact, the Sanger Institute, and Synthesia. It was particularly energizing to feel the energy of the ecosystem, which skews towards research thanks to excellent universities and research centers such as TUM, LMU, Max Planck and KIT. We’ll be back in Munich on 25 Feb 2025 and Berlin the week before on 20 Feb 2025 - register here! We were in Paris next, for the first of three events, including holding two Paris AI meet-ups and an apéro over at Station F. As usual, the scene was buzzing with enthusiasm as the French technology ecosystem was surfing a swell of good vibes in early 2024. We revved up our events in the US and held 11 different events in New York and SF over the course of the year. These included AI meetups, dinners and happy hours. We hosted speakers from Google DeepMind, Meta, ElevenLabs, Intenseye, Palantir, Replit, Captions, and NVIDIA. We’ve been particularly bullish on the talent base in New York and have made two new investments there in 2024. June saw the return of our annual flagship AI conference, @raais, which brought together 200 AI-native researchers, entrepreneurs, and operators for a day of talks, networking, and best practice sharing. We had a great range of speakers from industry and academia, touching on subjects in AI-first biology, embodied AI, design, coding, video, generative worlds and more! You can catch all the talks on our YouTube channel and our takeaways in the August edition of Guide to AI. We’ll be back with the 8th annual RAAIS conference on 13 June 2025 with another stellar line-up. Please register your interest at raais(dot)co All proceeds from RAAIS go towards the RAAIS Foundation, a UK-registered charity that aims to broaden the range of people who participate in advanced AI research and to make educational content freely available to the public. With the support of our sponsors, we relaunched our RAAIS Fellowships, which provide cloud credits and money to researchers working on open source AI-first projects, who don’t have the support of big labs. We’re starting to go through applications, but it’s not too late to apply. Meanwhile, we swung by Vienna in July for ICML, where Nathan and local portfolio founder Nikolaus Krall of Exscientia/Recursion hosted a (schnitzel, of course) dinner with research friends, old and new. And in November, we brought friends together from Delian Alliance Industries, Granola, Meta, and Odyssey for our latest London AI. This edition was particularly topical as we addressed the burgeoning research domain of open endedness for learning agents, autonomous defense systems, generative worlds, and product design principles for ever improving AI models. We’re building up another great schedule of events for 2025, which you can sign up to via our events page to get notifications. So far, you can join us in: - Berlin and Munich in February - New York in April, May and September - SF in April and then October for the State of AI Report launch - Paris in November … and more to come! HERE'S TO A BIG 2025 We started this round-up by saying that it’s still only day 1 for AI. We can’t wait to see what the community achieves in the coming year!

for the third year running, @spaniel_bashir and i chat on @gradientpub about the ai highlights of 2024 and what’s in store… 2 hour pod link below!

it’s the most wonderful time of the year (yearly AI roundup ft. @nathanbenaich)

during our 2-hour review of AI in 2024 @nathanbenaich and I spent a few minutes being China Guys........ some thoughts on the traditional narratives we have about US/European/Chinese innovation, reflections on an interview with the DeepSeek CEO translated by @jordanschneider and co

daniel and i spent 2 hours together yesterday discussing ai over the last year - reasoning, bio, chips, startups, vibe shifts, policy, and ofc wassup @airstreet :-) episode dropping soon on @gradientpub 💙

holidays are upon us again and for the third time @nathanbenaich and I did our year of AI in review... we also talked about figuring out how to be an investor when there's no playbook for the job

I spoke with @Philip_Goff about how panpsychism helps explain consciousness, hybrid cosmopsychism, and the "post-Galilean" science of consciousness he's recently advocated for

I loved writing and editing articles for The Gradient, but sadly this has become untenable time wise over the last few years... but The Gradient lives on! See this announcement.

We’re low on editorial bandwidth, so we’re making a few (hopefully temporary!) changes to our process at @gradientpub — I sat down with @hughbzhang and @andrey_kurenkov to discuss our history and where things stand thegradient.pub/podcasts/some-…

AI has the potential to transform real-world domains. But can AI actually improve outcomes in live interactions? We conducted the first large-scale intervention of a Human-AI Approach that has statistically significant positive learning gains w/ 900 tutors & 1,800 K12 students.

I spoke with @jacobandreas about - “Grounding” questions and language modeling - How he understands world models - His development as a scientist