The world needs more of this 🚀
I'm starting a podcast, "Confession," where I invite tech veterans to share their painful lessons.
We all connect through shared pain. Hear founders & CEOs like Satya Nadella (Microsoft), Dara Khosrowshahi (Uber), Luana Lopes Lara (Kalshi), and others share their stories and lessto Confession on...
✅ Spotify: open.spotify.com/show/6x2J7r97U…
✅ Youtube: youtube.com/watch?v=sd5spH…
✅ Apple: podcasts.apple.com/us/podcast/con…
@SatyaNadella @Dkhos @luanalopeslara
2
2
47
4,232
Eiso Kant retweeted
Yes. What we don’t know is so vast that we must approach Nature with humility and an open-minded willingness to learn by observation. Moving to hypotheses too early risks the blindness of confirmation bias.
Freeman Dyson's teaches definition of Science ✍️
The whole point of science is that most of it is uncertain. That's why science is exciting--because we don't know. Science is all about things we don't understand. The public, of course, imagines science is just a set of facts. s a process of exploring, which is always partial. We explore, and we find out things that we understand. We find out things we thought we understood were wrong. That's how it makes progress.
-- as mentioned in a 2014 interview
📷 Science Photo Library
0
1
17
1,712
Not one to usually repost a fund announcement but BCV and in particular Enrique Salem has been the best partner we could have wished for at Poolside. There are few people I would blindly trust, Enrique is one of them.
Make money. Have fun. Live with integrity. Everything else can change.
Fund XI. $1.6B total capital for those who know that it will.
2
3
100
6,299
Leaving image generation to diffusion but reasoning to an LLM is 🤩
Now take a step further and think about it in the context of robotics with spatial-action models.
Work combining modalities is still in the early days.
Now take a step further and think about it in the context of robotics with spatial-action models.
Work combining modalities is still in the early days.
8/ With that, we reframed multimodal generation as structured text/code generation. Diffusion just renders pixels. Planning, logic, reasoning all live in the LLM — so training looks like normal LLM training, and inherits all benefits of it: data + model scaling, reasoning, RL, tool use.
5
1
68
6,917
This.
“If it’s not painful you’re clearly not pushing hard enough.”
@travisk on entrepreneurship, pain, progress, and excellence:
4
2
73
7,884
Testing some new features (experiment API, improved component compatibility testing, new trajectory store) in the @poolsideai Model Factory and with a brand new UI! Just a few of our experiments are on it so far but looking promising.
5
4
104
5,059
Eiso Kant retweeted
Agentic evals are messy. A benchmark score tells you something about model performance, but it also reflects the whole system around it: the harness, sandbox, dependencies, timeouts and sometimes a loophole the agent found in the task.
That’s why trajectories matter so much to us. They show what the agent actually did and whether the score means what we think it does. We publish them to make that evidence transparent and auditable, giving the wider community more to learn from.
Watch @aalSonOfRavi and @ConnorBAdams go deep on all of this with @petergostev from @arena, including some surprisingly creative reward hacks!
That’s why trajectories matter so much to us. They show what the agent actually did and whether the score means what we think it does. We publish them to make that evidence transparent and auditable, giving the wider community more to learn from.
Watch @aalSonOfRavi and @ConnorBAdams go deep on all of this with @petergostev from @arena, including some surprisingly creative reward hacks!
Meet… Arena Conversations. @petergostev sits down with @poolsideai researchers,Connor Adams and Aalhad Patankar, to discuss how they’re building frontier open coding models.
Episode drops today at 10am PT on our YouTube.
They dig into Poolside's Laguna model family, why the jectories (not just benchmark scores) for anyone to audit, and how "experiments" at Poolside range from data mixes to harness design to reward-tuning decisions run tens of thousands of times a day.
2
8
69
9,542
Great work!!
We ran the largest open experiment on how frontier models do AI research.
100+ autonomous runs across 10+ models, sandboxed on 8xH200s for up to 8 days, iterating on the nanoGPT optimizer track.
Best runs closed 82% of the gap to a record built by dozens of humans over months.
1
0
33
4,277
Eiso Kant retweeted
Objects in the rearview mirror may be closer than they appear
Congrats to @Zai_org on GLM-5.3. It massively beats every American open model: Nemotron, Laguna, Inkling. It's not close.
GLM-5.3 uses the same base model as GLM-5.2; every gain came from post-training. If America keeps going down the "Nemotron Committee" path instead of ion, it will keep losing.
6
9
386
39,927
Eiso Kant retweeted
we knew the community could make Laguna faster. 2.6x faster is pretty fun :)
huge congrats to the winner, and to all 35 solvers who spent the last few weeks pushing Laguna XS 2.1 further!
more reasons to build in the open, together 🤝
huge congrats to the winner, and to all 35 solvers who spent the last few weeks pushing Laguna XS 2.1 further!
more reasons to build in the open, together 🤝
8
15
164
13,353
March 2024 (11 months after starting poolside) I woke up at 4am and wrote down 2 hours of thoughts on models and where LLMs should be going. Interesting to see how much of it has played out since.
Shared back then only with @PengmingWang, @NikolayZinov and @pmarca
1/🧵
Shared back then only with @PengmingWang, @NikolayZinov and @pmarca
1/🧵
13
5
103
8,992
