Nathan Benaich
@nathanbenaich
Followers
59K
Following
30K
Media
5K
Statuses
27K
solo member of investment staff @airstreet @airstreetpress @stateofaireport @raais
Joined November 2011
🪩The @stateofaireport 2024 has landed! 🪩. Our seventh installment is our biggest and most comprehensive yet, covering everything you *need* to know about research, industry, safety and politics. As ever, here's my director’s cut (+ video tutorial!) 🧵
31
312
1K
12 hours and counting - notary reads every single word of Series A docs in Germany out loud in front of founders. In person. Guys, we have GDP to grow here. Pure prehistoric madness.
358
419
9K
Current status in Soho.
“Nothing will ever be the same again”.
407
2K
4K
Meta Reality Labs is gonna cost as much as the Apollo Space Program, or about 10 Manhattan Projects. 🫠
113
469
3K
🪩The @stateofaireport 2023 is now here. Our 6th installment is one of the most exciting years I can remember. The #stateofai report covers everything you *need* to know, covering research, industry, safety and politics. There’s lots in there, so here’s my director’s cut 🧵
61
528
2K
🪩The @stateofaireport 2022 is live!🪩. In its 5th year, the #stateofai report condenses what you *need* to know in AI research, industry, safety, and politics. This open-access report is our contribution to the AI ecosystem. Here's my director's cut 🧵:.
27
484
2K
Interesting 👀 . “We are not going to stop at replicating ChatGPT. We want to build the assistant of the future, able to not only write email and cover letters, but do meaningful work, use APIs, dynamically research information, and be personalized.”
41
266
2K
The EU enters the large model training arena with. $56M. "This morning Brussels announced plans to develop an open source AI model of its own, with $56 million in funding to do it. The investment will fund top researchers from a handful of companies and universities across
245
94
2K
Prompt engineer is not a thing. Stop trying to make it a thing.
160
75
2K
2023: agi has the potential to destroy humanity as we know it . 2024: plz use my app
32
81
2K
It's not surprising that Jasper (vertical SaaS powered by GPT3) makes more $$$ than OpenAI (GPT3 API). A model behind an API with standard pricing basically treats every API use case as having the same value. This means the model host suffers suboptimal value capture. /n.
34
121
1K
Meta’s LLM for software testing work is super exciting. This paper describes Meta’s TestGen-LLM tool, which uses LLMs to automatically improve existing human-written tests. TestGen-LLM verifies that its generated test classes successfully clear a set of filters that assure
15
261
1K
When one of the most impressive contributors to AI says that 80% of his code is now written by AI…. …big thing are coming for the world of programming, dev tools, infra et al. Perhaps one doesn’t need to learn to code after all?.
Nice read on reverse engineering of GitHub Copilot 🪄. Copilot has dramatically accelerated my coding, it's hard to imagine going back to "manual coding". Still learning to use it but it already writes ~80% of my code, ~80% accuracy. I don't even really code, I prompt. & edit.
51
183
1K
You know someone’s lost the plot when they say things like “it’s nothing revolutionary” to a product that’s caught lightning in a bottle. That’s not the point. The point is that tons of people find it actually useful. Shame to see.
161
62
1K
🤓In 2017, Google researchers introduced the Transformer in "Attention is all you need", which took AI by storm. 5 startups were born: @AdeptAILabs (🏦 @airstreet), Inceptive, @NEARProtocol, @CohereAI, CharacterAI. Only 1/8 authors remain @GoogleAI, another is at @OpenAI. 😉
17
180
1K
"The idea that artificial intelligence could lead to the extinction of humanity is a lie being promulgated by big tech in the hope of triggering heavy regulation that would shut down competition in the AI market, one of the world’s top AI experts warned." - Andrew Ng.
35
249
1K
bye bye a bunch of startups 😔.
We've just launched fine-tuning for GPT-3.5 Turbo! Fine-tuning lets you train the model on your company's data and run it at scale. Early tests have shown that fine-tuned GPT-3.5 Turbo can match or exceed GPT-4 on narrow tasks:
44
87
1K
From DeepMind to Google DeepMind to DeepMind to Google DeepMind over 13 years :-).
27
80
1K
nov 30, 2022. we’re launching chatgpt today . don’t expect a significant impact on sales
9
51
980
Giga-GAN based upsampling. I mean these results are just mind blowing compared to where we were in 2015/16 with super-resolution.
20
110
831
When actual finance guys get into AI infra, they think of actually smart ways to finance their growth re: don't sell equity for GPUs. "We negotiated with them to find a schedule for how much collateral to go into it, what the depreciation schedule was going to be versus the
21
68
748
🇪🇺European VC at Series A: “can you forecast your paid marketing spend per channel so we can understand your CAC evolution”. 🇺🇸US VC at Series A: “what’s it going to take to crush your competition and be #1 in your category”. You pick :-).
33
52
750
💥We're live with @stateofaireport 2020💥. For the 3rd year, @soundboy and I spent 2 months analyzing the most interesting developments in AI. We aim to trigger an informed conversation about the #stateofai and its implication for the future. Thread 👇.
21
302
732
Bye bye diffusion, hello latte art!. This is impressive, fast, and efficient. Muse is a text-to-image Transformer model that achieves SOTA image generation performance while being significantly more efficient than diffusion or autoregressive models.
10
93
708
This is what I'm here for:. We examine breakthroughs over the past decade that include self-supervised learning, which allows models to be trained on vast amounts of unlabelled data, and geometric deep learning, which leverages knowledge about the structure of scientific data to
10
91
657
👋 I'm excited to unveil @airstreet’s second fund of $121,212,121 as we accelerate our mission to back ambitious AI-first companies in North America and Europe!. 🧵 My reflections on the journey, opportunity and what this means for our founders and community:
88
52
684
The @stateofaireport 2021 is live! . Est. 2018, @soundboy and I compile the most important work in AI research, industry, talent, and politics to inform conversation about the #stateofai. Our report is open-access to all. Here's a director's cut 🧵:.
14
235
578
Can Meta just please become an AI for science company? . Between this and all the awesome AI for proteins, materials, climate, medicine…Meta’s AI research contributions are so cool. Imagine what $250b would do!. I would support the pivot to MetAI.
🪐 Introducing Galactica. A large language model for science. Can summarize academic literature, solve math problems, generate Wiki articles, write scientific code, annotate molecules and proteins, and more. Explore and get weights:
18
51
555
ChatGPT is way more impressive and powerful than text-to-image/video models. There, I said it.
32
46
540
EIGHTY SIX THOUSAND EUROS. that’s how much this german notary ordeal taxed (sorry, cost) this series a startup to complete their financing . oh, and lawyer fees not included. um gottes willen 🤯.
12 hours and counting - notary reads every single word of Series A docs in Germany out loud in front of founders. In person. Guys, we have GDP to grow here. Pure prehistoric madness.
38
35
499
mamma mia! this approach and vision was what @twentybn friends were building a few years ago. oh what a few years does to trends! cc @muellerfreitag. "Li’s vision for spatial intelligence is even more ambitious: training a machine capable of understanding the complex physical
9
50
480
💥Welcome to @airstreet! We've raised a $17M fund to back early-stage “AI-native” founders solving major problems in the enterprise, consumer and life sciences sectors. Learn more about us in @jeremyakahn's @FortuneMagazine piece and our blog post below!.
73
60
455
alphafold.
what is your favorite source of protein?.
14
25
452
Open source is one of the biggest drivers of progress in software - AI would be unrecognizable without it. However, it is under existential threat from both regulation and well-funded lobby groups. The community needs to defend it vigorously. 🧵
37
108
385
A good piece by @aaronpholmes that provides some evidence for the following motion I’ve been describing to those who ask about defensibility:. - prototype with a 3rd party paid model API (ie OpenAI).- iterate to PMF.- get enough data to understand the economics (cost vs revenue)
30
48
429
In May 2021, I wrote an @FT oped to catalyse a discussion on rewriting the university spinout playbook. We've since crowdsourced deal terms from spinouts all over the world. Today, we’re sharing the @spinoutfyi database: The largest of its kind �.
18
132
416
you just gotta love this website
I am deeply pleased by this result, after ~72 very intense hours of work. Coming into OpenAI, I wasn’t sure what the right path would be. This was the pathway that maximized safety alongside doing right by all stakeholders involved. I’m glad to have been a part of the solution.
8
15
391
Today’s review marks a huge step forward for UK spinout founders 🇬🇧. It adopts almost all of the policy reform proposals I made via @spinoutfyi and have been pushing on for 2 years. A few thoughts on why this is a big day for the ecosystem and UK science&tech below:
36
80
402
Kids are banned from using generative AI at school. Meanwhile, their parents depend on it at work.
43
48
381
Our empirical findings suggest that Transformers solve compositional tasks by reducing multi-step compositional reasoning into linearized subgraph matching, without necessarily developing systematic problem- solving skills. To round off our empirical study, we provide
4
59
385
"We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.". Understatement of the decade, 12 June 2017.
6
51
364
hot take: deepseek is a blessing for ai startups that can now rip out their costly americano models and have profitable unit economics . their investors should be happy, happy as a hippo
27
30
355
Generative AI is becoming pretty astonishing. Prompt: epic sunset with blue and orange hues, and a beach with a surfer taking photographs with another surfer next to them next to a pier.
31
23
345
forgot to add that the notary fee is a % of the transaction size :-).
12 hours and counting - notary reads every single word of Series A docs in Germany out loud in front of founders. In person. Guys, we have GDP to grow here. Pure prehistoric madness.
19
9
342
Today, EU+UK universities have the ingredients to be spinout powerhouses: world-leading research, ambitious academics, and VC funding. Instead, our unis propagate an unproductive downward spiral of short-termism that hamstrings founders. Thread 🔽.
10
127
332
LLMs outperform RL at game play by studying papers and reasoning through chain-of-thought. So sick. LLMs could be the native language to teach robots. More soon!
20
71
337
🕵️♂️ State of AI Report 2018 - hot off the press @soundboy and I have set out to capture a snapshot of the exponential progress in AI with a focus on developments in the past 12 months. It's built for everyone, no expertise required. Check it out!
11
152
327
Every company has reason to be, but I can’t help but feel that sometimes capital allocation is messed up:.- VCs are going mental on 10 min grocery delivery: $290m raised by 1 company today. - Meanwhile, we announced $6m to give every cancer patient a fighting chance at survival.
17
61
315
EU announces supercomputer access to companies developing AI and will run a competition to select 4 companies to provide 4 million hours of supercomputer time to train large-scale AI. So that's llama-2 in 3 flavors, trained once, for the EU. I want to cry.
29
33
307
"From a scientific perspective, the robustness of these LLMs to the deletion of layers implies either that current pretraining methods are not properly leveraging the parameters in the deeper layers of the network or that the shallow layers play a critical role in storing
17
45
318
Rumour has it that VCs cancelled holiday plans because of ChatGPT and AI startups. Happy holidays!.
14
21
318
In October ‘23, @stateofaireport reached its 5th anniversary. With 934 slides over 6 editions serving as a historical document, we’re opening 2024 with a retrospective of the main storylines, what we predicted, and what we didn't. Welcome to the State of the State of AI. 🧵
8
91
308
Coming out of the craziest weekend in tech since SVB: .- MSFT in a stronger position than ever w/Sama+Greg+loyalists,.- OAI puts the breaks on shipping models and will likely lose lots of talent,.- New AI startups will form .- Threat of “OAI has won” is over. - GOOG vs. MSFT now.
13
24
305
Sounds like Jasper might be moving away from relying on OpenAI’s API by building their own models and training on Cerebras?. If so, interesting! Start out w/3rd party model API to go fast, build market share, then dive down the stack to protect yourself.
16
28
308
New York 🗽 will overtake San Francisco as a technology epicenter for the US. The 🍎 magnet is becoming stronger and stronger.
27
20
285
llm routing systems - which decide which llm to use for what task - are getting more important, esp. if open source keeps flourishing. yet there's no standard evaluation method for routing systems. enter ROUTERBENCH. As the range of applications for Large Language Models (LLMs)
9
42
282
News is out: Super excited to join the incredible @PointNineCap team as Venture Partner! Big things on the horizon. 🚀
35
17
279
this is the key point - if oss means you set the standard and the industry adopts it, you benefit tremendously.
7
25
267
🌟 @airstreet update! 6 months after public launch our $17M Fund has upscaled to $26M!. The market to build AI-first tech and life science companies to solve huge global challenges is massive. Our largest investor is @PrimaMateria_ led by @shak @eldsjal @brettbivens! Thread 🔽
27
18
276
🤓 Here it is, the 2019 State of AI Report by @soundboy and I. 130 slides covering the most important machine learning research, industry and political developments over the past 12 months. New section on China. 👉 Please RT if you find interesting!
5
167
253
🤖If LLMs are the foundation for AI, we have a lot to dev and learn from one another. 🤓I'm thinking of running a little sesh (IRL or virtual) to share ideas, tips/tricks, fav tools, design patterns, an open questions w/AI founders + builders. 👋Interested?.
80
13
253
this is so beyond bananas - eu policy wonks losing their marbles
24
23
248
Today is AI Demo Day:.- Adept $350M round.- OpenAI GPT-4.- Anthropic Claude.- Google PALM API.- UK Foundation Models task force.
5
37
249
Huge from our friends at @ProfluentBio! . Introducing OpenCRISPR: the world’s first ever open-source AI-generated gene editor. This is a huge milestone in personalised medicine.
4
48
251
The algorithm of Elon:. It had five commandments:. 1. Question every requirement. Each should come with the name of the person who made it. You should never accept that a requirement came from a department, such as from“the legal department" or "the safety department." You need.
6
37
241
In preparation for England vs Italy, if you have even a trace of pineapple in your Italian restaurant, watch out. 👀
9
57
234
after a few hours of vibe coding, i stand corrected, that it's actually not great just yet for these reasons:.- even when you're specific about what you want (eg clean this/that), it makes mistakes and if you don't know what you're looking at, you accumulate bugs and tech debt.-.
56
8
244
It's been a while now, but I wrote about this concept 3 years ago: . "I argue that the best way to capture value in ML is to directly monetize your predictions by being full-stack vs licensing ML tools as software.".
12
13
238
🙏Question to ML friends: How do you go about version control for your ML projects (data, models, and intermediate steps in your data pipelines)? Have you built your own tools? Are using something open source? Or a SaaS? Or does this come bundled with your ML infra products? Thx!.
59
57
233
🧬Today is a big day for AI-first biology!. 🤓@thisismadani et al in @NatureBiotech: LLMs learn to generate protein sequences with a predictable function across large protein families. 🆕@ProfluentBio launches w/$9M from @airstreet @insightpartners!. 🧵🔽.
5
35
228
Why does Verily, a biotech company, publish AI research on a fundamental computer vision problem with tasks totally unrelated to bio? . Seems odd? .
35
13
228
👋 @elonmusk - the A100 and H100 data we show is publicly disclosed information by the respective organisations. More than happy to add updated numbers for @Tesla @xai @X if you share them. I agree - from what I heard, they're very large :-). But we prefer not to leak secrets.
@thetechbrother This is not accurate. Tesla would be second highest and X/xAI would be third if measured correctly.
6
30
228
“OpenAI may try to raise as much as $100 billion in the coming years to achieve its aim of developing artificial general intelligence that is advanced enough to improve its own capabilities”. I mean, at this point, who is counting.
28
39
218
“The new spring in artificial intelligence is the most significant development in computing in my lifetime”.- Sergey Brin
6
36
224
in 2024, i wish for scientific literature paywalls to come down. it is an unnecessary tax on human knowledge.
11
21
214
overheard in the wv:. guy to girl: “i work in vc”. girl: “cool - so how do you decide which company to invest into?”. guy: “vibes”. girl: “surely not”. guy: “surely yes.”. vc is doomed.
35
4
218
Here are our 10 predictions for the next 12 months! Covering:. - GenAI/film-making.- AI and elections.- Self-improving agents.- The return of IPOs.- $1 billion+ models.- Competition investigations.- Global governance.- Banks + GPUs.- Music.- Chip acquisitions
7
20
212
the fine folks @huggingface have just recently published their guide to building 🍷FineWeb, a fully-open source training dataset for llms. it makes for a fun and educational read. thank you @thom_wolf and team
1
31
197
🌱There is now a non-negligible number of talented product and engineering folks in my network from unicorn or high-potential startups who have quit to explore or actively build companies focused on mitigating climate change. Let's discuss it! 🌎.
22
23
202
From all the YC companies in this last batch, apparently Oxford is the #2 most common Alma Mater. Unsurprisingly, Stanford is #1. Huge opportunity for @UniofOxford to really lean into building a legit current student and alumni community like we have in the US.
16
28
208
Companies born in our universities need more than government spin. In the @thetimes, I argue that rewriting the playbook for creating companies from uni inventions is critical to tech sovereignty. Britain won't be a Science Superpower w/out spinouts.
9
62
207