Alexander Leirvåg retweeted
Introducing Trinity-1, the first interactive gaussian avatar, available in preview via our API today. Priced at less than 1 cent per minute, we are making interactive avatars finally viable for real world use cases. Try our demo and create your own gaussian avatar.
Alexander Leirvåg retweeted
Being good at your job ≠ being good at the internet.
If you’ve ever felt like “visibility” is a scam, same.
We’re building a tool for people who want to get seen without selling their soul to the algorithm.
Take this 2-minute survey to help use validate and get to know your needs.
You’ll get:
• Early access
• Your visibility vibe decoded
• Ways to attract better opportunities without posting online!
• (And maybe learn something about yourself in the process)
Start here →
docs.google.com/forms/d/e/1F…
Alexander Leirvåg retweeted
Replying to @francoisfleuret
Alexander Leirvåg retweeted
Hiring is a sign of failure.
The perfect company is a computer plugged into a wall.
100 employees means 100 roles that you failed to automate.
Hiring should only be done as a last resort.
Alexander Leirvåg retweeted
AlphaFold 3 is now open-sourced (without gatekeeper).
The revolutionary significance cannot be overstated.
It's not a fight, some drama, or a technology you can see with your eyes, so it get's very little attention.
However don't be fooled by what others are or are not paying attention to. What the masses pay attention to in any given moment has little to do with what history will value with the perspective of decades or centuries. I think the future will single out AlpaFold.
Here's the context:
While DNA is the code of life, proteins are its substrate, the hardware necessary to give the code any meaning.
Biotechnology minus the understanding and control of protein was doomed to a deadlock, like computer science before the Turing machine.
We now know how the linear code works, moving from linear to 3D, from the instructions to function was the missing piece.
The ability to predict structures, and generate novel matching proteins, will give birth to a new era of biological capability in human vitality, materials, energy, environment, and countless other areas.
Well done @demishassabis, John Jumper, and David Baker.
Alexander Leirvåg retweeted
“The doers are the major thinkers.
The people who really create the things that change this industry are both the thinker-doer in one person.”
— Steve Jobs
Alexander Leirvåg retweeted
Peter Thiel:
“[…] if you hear the words ‘big data’ and ‘cloud computing’, you need to run away as fast as you possibly can. Just think ‘fraud’ and run away.”
Alexander Leirvåg retweeted
partial list of things big labs use but do not share details about:
pretraining data composition
synthetic data generation pipelines
RL{AI,H}F objectives
"Reasoning"
distillation
test time compute scaling
this is also probably a list of things you should be thinking about
Alexander Leirvåg retweeted
The overlooked GenAI use case: cleaning, processing, and analyzing data.
blog.sumble.com/the-overlook…
Job post data tell us what companies plan to do with GenAI. The most common use case is data analytics projects. Examples:
- AstraZeneca: using LLMs on freeform documents to structure results from their Extractables & Leachables testing (sumble.com/l/FgehjrgnvN)
- Trafigura: The Document AI team is using LLMs to extract data from a corpus of commodity trading documents to generate credit reports
(sumble.com/l/6bY7mhyAHd)
The startup ecosystem is overlooking this use case, instead focusing on other areas such as customer support, sales & marketing and code gen.
Alexander Leirvåg retweeted
why do language models think 9.11 > 9.9?
at @transluceAI we stumbled upon a surprisingly simple explanation - and a bugfix that doesn't use any re-training or prompting.
turns out, it's about months, dates, September 11th, and... the Bible?
Monitor: An Observability Interface for Language Models
Research report: transluce.org/observability-…
Live interface: monitor.transluce.org/ (optimized for desktop)
Alexander Leirvåg retweeted
I was rewriting the Entropix equations into math notation (per head), and I find it interesting that they define logits varentropy as the variance of the Shannon information content while the attention varentropy is defined as the variance of the attention entropy
Alexander Leirvåg retweeted
Google Deepmind trained a grandmaster-level transformer chess player that achieves 2895 ELO,
even on chess puzzles it has never seen before,
with zero planning,
by only predicting the next best move,
if a guy told you "llms don't work on unseen data", just walk away
Alexander Leirvåg retweeted
Founders should be careful of spending too much time around VCs when ideating or figuring out to do next.
Most VCs drastically over-intellectualize company building while underrating how much of winning isn't strategy but brute force warfare. Hand-to-hand combat.
Especially true if the investor has never built something before themselves.
And being an APM at Google does not count as having built something.
Finding the gap in a market map want yield a great product, happy customers, insights, or unfair advantages.
There is value you can get from talking to select investors. But be careful of the weight of time spent here.
Alexander Leirvåg retweeted
introducing "ditto"
the simplest self-building coding agent
📄 ~500 lines of code
🛠️ can build multi-file apps
🔁 a simple LLM loop with 5 tools
github/replit/more below 👇
Alexander Leirvåg retweeted
Nvidia’s Nemotron 70B is destroying GPT4o & Sonnet 3.5 on Arena Hard, MT-Bench, AlpacaEval 2 🔥
Arena Hard
Nemotron: 85.0
3.5: 79.2
4o: 79.3
AlpacaEval 2:
Nemotron: 57.6
3.5: 52.4
4o: 57.5
MT Bench
Nemotron: 8.98
3.5: 8.81
4o: 8.74
LFG!! 🔥🔥
huggingface.co/collections/n…



























