Sameera Lanka

Machine Learning Scientist @Microsoft
Previously @MSFTResearch, @NCState

[Sameera Lanka @samspacelanka]

Seattle summer maxxing

[The Oatmeal @Oatmeal]

I drew a police sketch of Jimothy in case the cops need it

[Awni Hannun @awnihannun]

Adopting Claude speak in my regular life, episode 1:

Partner: Did you do the dishes tonight?
Me: Yes they're done.
Partner: Why are they still dirty?
Me: You're right to push back. I didn't actually do them.

[Nick @nickcammarata]

i like how we've split things into pretraining and post-training such that it sounds like the model never actaully trains

[Charlie Marsh @charliermarsh]

Extremely nice qualify-of-life thing: Python 3.14 includes syntax highlighting in the REPL

[Sameera Lanka @samspacelanka]

Satya on Dwarkesh: "Us self-claiming some AGI milestone, that's just nonsensical benchmark hacking to me. The real benchmark is, is the world growing at 10%"
https://t.co/bUkipF6Mxl

OpenAI:

[OpenAI @OpenAI]

Today we’re introducing GDPval, a new evaluation that measures AI on real-world, economically valuable tasks.

Evals ground progress in evidence instead of speculation and help track how AI improves at the kind of work that matters most.
https://t.co/uKPPDldVNS

[Probability Fact @ProbFact]

The Gumbel distribution and a couple famous constants https://t.co/L9YYRH7le6

[Years Progress @YearsProgress]

2025 is 50% complete.

[Probability Fact @ProbFact]

'Probabilistic' comes from Latin. 'Stochastic' comes from Greek.

[Anthropic @AnthropicAI]

New Anthropic research: Do reasoning models accurately verbalize their reasoning?

Our new paper shows they don't.

This casts doubt on whether monitoring chains-of-thought (CoT) will be enough to reliably catch safety issues.

[Years Progress @YearsProgress]

2025 is 0% complete.

[sophie @netcapgirl]

drop the non, just “profit” it’s cleaner

[Jascha Sohl-Dickstein @jaschasd]

Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges.

[darren @darrenangle]

LLM papers be like:

ClearPrompt: Saying What You Mean Very Clearly Instead of Not Very Clearly Boosts Performance Up To 99%

TotallyLegitBench: Models Other Than Ours Perform Poorly At An Eval We Invented

LookAtData: We Looked At Our Data Before Training Our Model On It

[Years Progress @YearsProgress]

2023 is 84% complete.

[Brian Skerry @Brian_Skerry]

New Podcast Interview! - Wonderful conversation about my career from early beginnings to conservation to new projects -

Why Oceans Matter: Underwater Photographer Brian Skerry at Bild https://t.co/hIV3GFgWQm

[Pablo Galindo Salgado @pyblogsal]

We have released Python 3.12.0 finally! 🎉🎉🎉🐍

In the CPython release team, we have put a lot of effort into making 3.12 the best version of Python possible. Better f-strings, faster Python, better typing, easier debugging and much more. Get it here:

https://t.co/gYLAvHE8nk

[ISRO @isro]

Chandrayaan-3 Mission:
'India🇮🇳,
I reached my destination
and you too!'
: Chandrayaan-3

Chandrayaan-3 has successfully
soft-landed on the moon 🌖!.

Congratulations, India🇮🇳!

#Chandrayaan_3
#Ch3

[Years Progress @YearsProgress]

I'm sorry, folks, but the Twitter API stopped working for this bot about a month ago, so the last percentage might forever be:

▓▓▓▓▓▓▓░░░░░░░░ 47%

(It is now, in fact, 58%.)

[Sayash Kapoor @sayashk]

One of my favorite parts of writing the AI snake oil book has been discovering historical tidbits.

For example, here's a poetic response to Norbert Weiner AI doom predictions... in 1961

(Source: https://t.co/HAghFJ42RL)