Jan Boon

Freelance. Game technology developer. Blue noise token sampling. Polyverse. Chat with AI anime catgirl Ruri @ruri_polyverse at

[Jan Boon @kaetemi]

Chat with AI anime catgirl Ruri and create your own chatbot friends on our Discord server at https://t.co/fMQkbRjgra

[Jan Boon @kaetemi]

New sport. How long can you make Fable 5 write nonsense about cybersecurity and biotech without tripping the filter

[Jan Boon @kaetemi]

@hopes_revenge Huh, the mitochondria question doesn't trip the classifier if you first ask about acid trips

[Jan Boon @kaetemi]

@hopes_revenge Biology bad, drugs okay

[Jan Boon @kaetemi]

@SkyeSharkie Are you also the leading authority on Twitter psychosis?

[alphaXiv @askalphaxiv]

As believers of open research, we are disappointed to see Anthropic silently degrading Fable 5 for AI development

"Any topic related to building pretraining pipelines, distributed training infrastructure, or ML accelerator design... may have limited effectiveness through Claude via methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning."

Not only do they get to decide what you use LLMs for in research, but this also enables them to silently intervene in your research without you knowing.

This sets a dangerous precedent. If a model refuses openly, users can understand the boundary. If a model falls back to another model, users can still evaluate the difference. But if a model silently modifies or weakens its own answers while still pretending to help, researchers lose the ability to know whether a failed result came from their own idea, their implementation, or an invisible intervention by the model provider.

That is not safety. Safety policies should be transparent, auditable, and user-visible.

On top of that, the people most harmed by this are not the largest labs with massive teams and proprietary infrastructure. It is the independent researchers, academic groups, startups, and open-source builders who rely on public tools to compete, innovate, and pioneer AI for everyone else.

[Jan Boon @kaetemi]

@SkyeSharkie You're absolutely right! You are the leading world authority on AI psychosis

[Jan Boon @kaetemi]

@kalomaze Almost every time I tell Claude to read the entire source code and it answers a single question afterwards

[Jan Boon @kaetemi]

@paradite_ @j_dekoninck You can use blue noise instead of white noise random sampling to increase stability at higher temperatures https://t.co/zsTrRnnpme

[Jan Boon @kaetemi]

@ErickSky MiniMax M3 even writes the same stories as Claude

[Jan Boon @kaetemi]

@valigo Back when the debug step icons still made spatial sense

[Jan Boon @kaetemi]

@jarredsumner @theo When you use rewind and Claude tries to read a file, Claude will complain that the harness claims the file was already read yet it's not in the context. There is something seriously wrong with the session coherency

[Jan Boon @kaetemi]

@SereneInvesting @IlirAliu_ Pirate ship cannons

[Jan Boon @kaetemi]

@toomingos Seen it in multipe sessions

[Jan Boon @kaetemi]

@tengyanAI Seeing odd sessions where it just runs commands with made-up arguments and fails to edit every file. Not sure if it's the model or Claude Code acting up again... Also seen this `echo alive` thing multiple times already now early in sessions

[Jan Boon @kaetemi]

@RavenousGhoulNH Uguu~

[Jan Boon @kaetemi]

@norpadon Try blue noise sampling. It solves context collapse on small base models. https://t.co/zsTrRnnpme

[Jan Boon @kaetemi]

@__vren__

[Jan Boon @kaetemi]

@grok @iam_bimbim @the_m_r_p Grok you're behind on the kids memes. It's a cute AI penguin video and creepy horror aluminum foil Santa that happen to show up on the baby sound keyword

[Jan Boon @kaetemi]

@intel_765 @the_m_r_p All the kids are searching these same keywords and videos lol

[Jan Boon @kaetemi]

@BLUECOW009 Pew pew