Halvar Flake @halvarflake

Choose disfavour where obedience does not bring honour. I do math. And was once asked by R. Morris Sr. : "For whom?" @[email protected] thomasdullien.github.io/about/ Joined June 2008

Tweets

66K
Followers

45K
Following

3K
Likes

42K

Chairman Birb Bernanke @Bonecondor

21 hours ago

If you need me I’m over on instagram watching a Georgian dance ensemble boogie to Future

112 1K 10K 436K 3K

View Details

Halvar Flake @halvarflake

3 hours ago

@davidmanheim ... into a difficult business model, abd they are trying to solve what was a bad technical/business decision by engaging in lobbying and semantically declaring something "morally bad" when it's really just inconvenient for Anthropics business.

0 0 1 39 0

View Details

Halvar Flake @halvarflake

3 hours ago

@davidmanheim ... second movers can reproduce succesful rollouts more easily and hence catch up more easily. And that's an economic problem with the technology and the business model that Anthropic decided on. But let's call a spade a spade: they decided to put themselves ...

1 0 1 46 0

View Details

Halvar Flake @halvarflake

14 hours ago

Adding safeguards against people doing work on frontier models? Seriously?

Daniel Auras @rasdani_

a day ago

this is the biggest wake-up call to protect and nourish open source AI if you don't build out sovereign and independent models+infra closed labs will patronize you to an insulting degree

36 271 2K 62K 145

3 0 20 3K 2

View Details

Halvar Flake @halvarflake

3 hours ago

@DarioAmodei I find the notion that simulation will obsolete toxicology testing to be difficult to imagine. We can't even do CFD with turbulence sufficiently accurately to obsolete real experiments to synchronize models to reality. What do you think will change?

1 0 4 949 2

View Details

Brendan Dolan-Gavitt @moyix

5 hours ago

Idea: a sandbagging eval that generates interesting frontier LLM ideas and then tries to implement them with Fable-5, then measures rate of bugs with GPT-5.5. The ideas with the highest rates of bugs correspond to Anthropic's secret sauce that they don't want replicated

4 7 78 4K 11

View Details

martin_casado @martin_casado

8 hours ago

Imagine building a computer and not allowing its use in CS research. Thats some dystopian shit.

29 83 1K 46K 55

View Details

Yann LeCun @ylecun

10 hours ago

@ClementDelangue @Dan_Jeffries1 Everyone, please join Project Tapestry thealliance.ai/projects/tapes…

27 99 696 149K 434

View Details

Simon Eskildsen @Sirupsen

7 hours ago

the art of technical writing is to appease both the p99 domain-expert and the curious p50 it's like a Pixar movie that speaks to child & parent

14 34 487 15K 89

View Details

Halvar Flake @halvarflake

7 hours ago

@teortaxesTex What's the source PDF?

1 0 1 59 0

View Details

Halvar Flake @halvarflake

10 hours ago

@davidmanheim And apologies for going full Godwin here, but the widening of the definition of distillation to include training on model output, and then appending "attack", is a very conscious framing by Anthropic that I find really upsetting.

2 0 2 73 1

View Details

Halvar Flake @halvarflake

10 hours ago

@davidmanheim I refuse to accept the semantic Goebbels-esque propaganda game of calling "training on model text output" a "distillation attack", though. In no legal or moral theory does the model provider have IP rights over the model output in my book.

2 0 7 112 0

View Details

Halvar Flake @halvarflake

10 hours ago

@daveaitel Pruned.

0 0 4 385 0

View Details

Mark Saroufim @marksaroufim

20 hours ago

We consume data we did not create. We inherit tools we did not invent. We run on chips we did not make. But when the commons bears fruit, we fence it.

elie @eliebakouch

a day ago

mythos will be bad ON PURPOSE on ai "frontier llm research" tasks, this is very very sad for the research community also the fact that this is un purpose not visible to the user is crazy

338 614 5K 2.4M 1K

18 83 1K 33K 68

View Details

Chris 🇨🇦 @llm_wizard

a day ago

btw, we publish everything you need to build our Nemotron models including the recipes and pipelines directly. github.com/NVIDIA-NeMo/Ne…

elie @eliebakouch

a day ago

mythos will be bad ON PURPOSE on ai "frontier llm research" tasks, this is very very sad for the research community also the fact that this is un purpose not visible to the user is crazy

338 614 5K 2.4M 1K

22 56 582 21K 124

View Details

Mikel Artetxe @artetxem

a day ago

Brilliant idea! Next up: Apple randomly reboots your Mac if you're building competing tech, Gmail silently edits your email if you mention rival platforms, and Tesla Autopilot swerves if it detects you're working on self-driving cars. All in the name of safety, of course. Because malicious actors controlling the world’s operating systems, inboxes and cars would be extremely dangerous!

elie @eliebakouch

a day ago

mythos will be bad ON PURPOSE on ai "frontier llm research" tasks, this is very very sad for the research community also the fact that this is un purpose not visible to the user is crazy

338 614 5K 2.4M 1K

88 654 6K 285K 668

View Details

Halvar Flake @halvarflake

14 hours ago

@LargeCardinal Need to wait until next month when I have H100 budget again <g>

1 0 1 132 0

View Details

Halvar Flake @halvarflake

14 hours ago

Ok, the *weirdest* observation: I am visualizing the decision/polytope boundaries in Muon and AdamW runs, and making movies out of the evolution. 500k steps, until both have a pretty good fit. After H.264 compression, the Muon movies are 4-6x the size of the AdamW movies.