We just finished setting up the Exacluster:
- 144 H200s
- 3456 CPUs
- 270TB NVME SSD
- 20TB GPU RAM
- 36TB CPU RAM
- 100KW operating power
Prepare yourselves for what's coming..
Thoughts on the eve of AGI
I talked to several friends about o3 this week. Their summarized response is basically "holy crap is this actually happening?"
Yes, this is actually happening. The next few years are going to be insane. This is historic stuff, galactic even.
What's
this guy built our distributed ML training infrastructure from scratch
Yes, that is a tmux session for our gpu cluster, on an iphone, at a birthday party
This is the first time in history that you can download whatever comprehensive dataset you need from the web.
- Every ML engineer in SF
- Every sales person in NYC who has a technical background
- all international VC firms, plus their size, founding date, and industry focus
I spoke to / heard from a few senior AI labs people this weekend and I now feel even more confident in these predictions from a month ago.
Hearing these same ideas come out of their mouths was quite strange in how real it became. There is a sequence of future events that if you
Thoughts on the eve of AGI
I talked to several friends about o3 this week. Their summarized response is basically "holy crap is this actually happening?"
Yes, this is actually happening. The next few years are going to be insane. This is historic stuff, galactic even.
What's
o1 is the biggest AI update since transformers and it paves a clear path toward AGI.
This is what Ilya saw. This is why Sam wanted trillions of dollars of compute. This is what Leopold warned us about.
Just like transformers, o1 lets you efficiently trade in compute for
We shipped the slowest search engine in history, and that's why you should want it.
We built Exa Websets to get you a comprehensive list of whatever you want from the web. That necessarily takes time.
For every search, we deploy thousands of AI agents that recursively call Exa
It's honestly insane that LLMs haven't discovered something significant yet.
I believe this is a skill issue.
No one has tried to call these phd-level LLMs thousands of times with the right arxiv papers streaming in, prompted in all the right ways, chaos and randomness
Can AI discover new science?
Exa and Anthropic are cohosting the AI Nobel Prize Hackathon. Max out Claude, AWS credits, and MCPs to discover something new about the world.
In case your work doesn't win the Nobel Prize, we'll have prizes for the top 5 projects. Apply below👇
Spent the weekend hacking together Exa embeddings over 4500 NeurIPS 2024 papers - neurips.exa.ai
Let's you:
- do otherwise impossible searches ("transformer architectures inspired by neuroscience")
- explore a 2D t-SNE plot
- chat with Claude about multiple papers
We hosted a poker tournament for AI companies in the Exa office. Cocktail attire 🍸
Reps came from OpenAI, Deepmind, Anthropic, Nvidia, Databricks, Cursor, and more. Who do you think won?
We raised $85M. It's taken 4 years to get to this moment.
Exa began as a gnawing itch. What if Google was outdated? What if you could use transformers to build perfect search over all the world’s information? What if we were the ones who did it?
In 2021, this sounded crazy to
Everyone's building Deep Research. But no one's fixing the deepest problem with research...
You can't precisely find what you want from the web.
Today, we're opening up our Websets product which helps fix this.
Websets takes a complex query and finds a precise list of matching