Been making SF tech events my whole personality.
Now that @sambrunty is in town, it’s time we host our own. Hang with us and the good people at Daytona and Doubleword at our hackathon during #SFTechWeek.
1
3
49
RSVP to Model Training Night: partiful.com/e/zVfEE2R1zoj8W…
1
6
Fine-tuning a smaller open-weight model isn't a compromise.
For specialised tasks it can be the better way to build.
1
1
5
213
Across legal, biomedical and aviation benchmarks, our fine-tuned models were 7x more accurate at quoting clauses word for word and 20x cheaper on the legal task.
How we did it: overmindlab.ai/research/when…
1
6
136
A few years ago, you could mumble “we’re AI for [insert vertical]” into your third Guinness, and be reasonably confident that an investor a few tables over would lock eyes and invite you to pitch.
Today that’s like saying “my website is on the internet.”
1
1
6
37
If data is a by-product of intelligence, why not let your agent's intelligence produce its own?
Your @langfuse traces are already training data.
Import them into Overmind and turn selected runs into datasets for evals and training. Or send OpenTelemetry traces directly.
Start with the data your agent already produces.
ALT Langfuse import and direct OTLP export feed traces into Overmind for datasets, evals and training.
22
I’ve always been a huge massive nerd (spent an embarrassing amount of time building mods for Minecraft rather than playing sports and drinking in the park), and I’ve always wanted to work in tech. But when my time at university was wrapping up, I didn’t feel a pull from the Googles and Metas of the world that I thought I would.
1
1
18
89
The people actually doing that seemed to be working in the British intelligence, which is why I started my career there instead of big tech. I’m glad I did, because the work I did there was super unique and super complex. Every day we solved problems under extreme constraints. Without off-the-shelf products and playbooks, we had to build things literally no one else had ever built before.
1
1
12
Whether they admit to it or not, OpenAI and Anthropic are geopolitical actors. Any decision they make (or any decision made for them, based on the country they’re headquartered in) ripples through every organization that’s built something on their APIs.
1
4
52
Reassuringly, the ecosystem already seems to be correcting for this. Open-weight models, Kimi, Qwen, all the rest, are hitting, if not superseding, frontier models.
1
2
14
Most agent evals just mark the final output which misses the point. The interesting signal is in how the agent got there.
Ask someone to grab me a latte and they respond by opening a coffee plantation. Eventually it arrives, the output is correct, but the process was absurd.
Mark the behaviours, not only the result.
ALT Text card: replace evals are vibes with a pinned eval set and a baseline score.
2
4
328
Open models are the future of AI. The facts are the facts.
Exciting day for NVIDIA and @huggingface.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. They allow every developer, startup, university, industry and country to build with, customize and benefit from AI.
Thank you @ClementDelangue for coming to me.
NVIDIA is going to be a great home for Hugging Face, its community and the future of open models. 🤗
blogs.nvidia.com/blog/nvidia…
1
5
101