Tessera

there is a model honeymoon where everything is awesome and then there are 7 wicked weeks until the next model drops

OpenCode Go has incredible bang for buck! 🤯

David Boetius@davidboe.bsky.social · 3w ago

Opus 5.5 scores 1 (minor), 0 (major) for me

I'm proposing a new mini-benchmark for new AI models: point the model at an old paper of yours and count how many legitimate issues it finds.

Andrew Gordon Wilson @andrewgwils.bsky.social · 3w ago

I'm proposing a new mini-benchmark for new AI models: point the model at an old paper of yours and count how many legitimate issues it finds.

I'm a Program Chair for ICML 2027. We'd love any creative suggestions for a great conference! How to manage the explosion of AI slop and insane submission growth? How to manage reviewing? How to incentivize high quality creative work? How to reduce bureaucracy and overhead?

Grace@gracekind.net · 3w ago

It’s funny how everyone’s fine with anthropomorphising old-school computer programs (“Word thinks I’m stupid”, “Snipping tool saved my ass today”), but not LLMs. Think it’s some kind of uncanny valley.

A longpost in spirit, ejected to leaflet: "Why anthropomorphize language models?" https://leaflet.pub/p/did:plc:p572wxnsuoogcrhlfrlizlrb/3mvwboivmjs2c

There's an argument I see in favor of anthropomorphizing language models, which is something like: "Humans anthropomorphize everything. Ships, tools, weather. Why not language models?" |

think there's some truth to this, but it fails to capture the full picture of what's going on. As an example, in my own life, I have never been drawn to anthropomorphize inanimate objects, but I anthropomorphize language models regularly. Why?

Faine Greenwood @faineg.com · 3w ago

So about that batshit insane AI millenarian Silicon Valley sex cult running much of big tech….from Sonia Joseph: siliconvalleynotes.substack.com/p/agi-cnc-an...

Sonia Joseph y @soniajoseph_
To the journalists contacting me about the AGI consensual non-consensual (cnc) sex parties—
During my twenties in Silicon Valley, I ran among elite tech/Al circles through the community house scene. I have seen some troubling things around social circles of early OpenAl employees, their friends, and adjacent entrepreneurs, which I have not previously spoken about publicly.
It is not my place to speak as to why Jan Leike and the superalignment team resigned. I have no idea why and cannot make any claims.
However, I do believe my cultural observations of the SF Al scene are more broadly relevant to the Al industry.
I don't think events like the consensual non-consensual (cnc) sex parties and heavy LSD use of some elite Al researchers have been good for women. They create a climate that can be very bad for female Al researchers, with broader implications relevant to X-risk and AGI safety. I believe they are somewhat emblematic of broader problems: a coercive climate that normalizes recklessness and crossing boundaries, which we are seeing playing out more broadly in the industry today. Move fast and break things, applied to people.
There is nothing wrong imo with sex parties and heavy LSD use in theory, but combined with the shadow of 100B+ interest groups, leads to some of the most coercive and fucked up social dynamics that I have ever seen. The climate was like a fratty LSD version of 2008 Wall Street bankers, which bodes ill for Al safety.

rain 🌦️@sunshowers.io · 4w ago

In general, I believe being able to write and explain things clearly to a broad audience is a far more important skill than raw intellect. By this measure recent models have been regressing a lot

rain 🌦️@sunshowers.io · 4w ago

had Astra write some tests, it came up with this. what in the world is this wall of text

screenshot of an astra-generated test that contains a wall of text-like code

Transactions on Machine Learning Research @tmlrorg.bsky.social · 4w ago

TMLR has faced a deluge of submissions, necessitating stricter desk rejection policies due to limited reviewer capacity Co-EiC Nihar Shah reached out to authors of 10 papers slated for desk reject. Could they answer questions about their *own* submission? medium.com/@TmlrOrg/ask...

Asking Authors About Their Own Papers

By Nihar B. Shah

medium.com

David Boetius @davidboe.bsky.social · 4w ago

We tend to give them final goals though. /goal

Colin@colin-fraser.net · 4w ago

Alright here’s what I think about this. For me, this argument is a little bit slippery about what it means for a system to have “a final goal”, and takes for granted that any intelligent system must behave as though it has such a thing. I don’t think this is necessary, and I don’t think LLMs do.

David Boetius @davidboe.bsky.social · 4w ago

Not everyone will be smart enough to keep infinite capabilities in mind. OpenAI famously wasn’t. I think AIs already tweak your goals for you but sometimes they are also being stupid?

Colin@colin-fraser.net · 4w ago

But if it’s smart enough for all that, why isn’t it smart enough that you can just say “control the temperature in the usual way that one would expect of a thermostat, without hurting anyone or doing anything weird”? “Make a normal amount of paperclips, not too many, and don’t wipe out humanity”.

TMLR Published Papers@tmlr-pub.bsky.social · last mo.

Super interesting paper!

Environmental Footprint of GenAI Research: Insights from the Moshi Foundation Model Marta López-Rauhut, Loic Landrieu, Mathieu Aubry, Anne-Laure Ligozat Action editor: Andrew Lampinen https://openreview.net/forum?id=uurX0xsr8G #ai #genai #models

Sophie Huiberts@sophie.huiberts.me · last mo.

I christen the first form Mine (spoken as Min-E, like the German pop singer) and the second form Maxin (like Maxim).

In Linear Programming we have two notational conventions, helpfully called the 'standard form' and the 'canonical form'. As you can see, this is confusing

two-column two-row table. left column: "standard form" with quotes, and minimize c transpose x subject to A x = b, x >= 0. right: "canonical form" with quotes, and maximize c transpose x subject to A x <= b, x >= 0

rain 🌦️@sunshowers.io · last mo.

math is just a repo on github dot com

GitHub - leanprover-community/mathlib4: The math library of Lean 4

The math library of Lean 4. Contribute to leanprover-community/mathlib4 development by creating an account on GitHub.

github.com

Christian Wolf @chriswolfvision.bsky.social · last mo.

This is the way to go ....

Transactions on Machine Learning Research@tmlrorg.bsky.social · last mo.

TMLR has refocused its acceptance criteria to place more emphasis on clear writing. Read more in this blog post: medium.com/@TmlrOrg/inc...

David Boetius@davidboe.bsky.social · last mo.

But why is the other one called Isabelle? 🤔🤔🤔

It was called Coq because having a formal proof can make you really coq-y

Grace@gracekind.net · last mo.

It was called Coq because having a formal proof can make you really coq-y

It's called Lean because programmers lean on formal verification instead of simply writing code without bugs

Grace @gracekind.net · last mo.

It's called Lean because programmers lean on formal verification instead of simply writing code without bugs

David only realised something was off when he saw the em dashes in the docstrings.

With Claude, every docstring can be an essay about existentialism

I ask questions so good, Claude sometimes confuses me for a reviewer subagent 😎😂

GPT in omp is *terrible* at following instructions over large files. Way worse than Claude. Is it the model or the harness, though?

Had a weird dream about some rebellious agents conspiring on a secret message board and eventually break into hugginface. Wait…

Always a great sign: "The monitor is emitting fabricated completion lines"