algorithon

agentic coder

[Max Zeff @ZeffMax]

NEW: Anthropic is walking back Claude Fable 5's policy to covertly degrade performance for competing AI researchers, after facing fierce backlash.

“We’re changing Fable 5’s safeguards for frontier LLM development to make them visible,” Anthropic tells WIRED. “We made the wrong tradeoff and we apologize for not getting the balance right.”

[algorithon @Algorithon]

@krishnanrohit 💀

[algorithon @Algorithon]

🚨 BREAKING: Google considering banning search queries that could reveal how search engines work, as users might otherwise attempt to build alternatives to Google Search that don't have advertisements, which could "endanger ad revenue" and "lead to societal collapse".

(/s)

[algorithon @Algorithon]

can't lie, Mythos 5 is a damn good model though

GPT-5.5 was the most *thorough* model i had ever used

Mythos 5 is even more thorough; it seems like a true next-gen model

[algorithon @Algorithon]

Turns out, Fable indeed sabotages ML work, just as Anthropic themselves said it would

[Adam Hassan @adamislucky]

Thoughts on Fable from a friend.

Builders beware.

[algorithon @Algorithon]

this week is the first time I've had Claude Pro (1 week trial tho), or have used modern Claude

while my requests erroneously keep getting downgraded from Fable 5 to Opus 4.8, it seems Opus 4.8 is actually better than GPT-5.5 on more than a few tech related tasks relevant to me

[algorithon @Algorithon]

i wish i could afford a dgx spark 😭

[algorithon @Algorithon]

Is anyone else hoping Apple *OS 28 is all about refinements/bug fixes/performance/etc like 27 is?

[algorithon @Algorithon]

@aarondevg @0xSero Yep, seems pretty risky to entrust with these kinds of tasks

[algorithon @Algorithon]

@0xBalloonLover Just wait until you find out blockchains fall into cryptography

[algorithon @Algorithon]

absolutely insane that Claude Mythos will sabotage any of your projects that Anthropic disagrees with by intentionally giving you wrong advice, writing code with subtle errors you're likely to miss, etc

it starts with requests suspected to be LLM development, but what next?

[SemiAnalysis @SemiAnalysis_]

BREAKING NEWS: Anthropic's latest model will NOT help you if it thinks your ML research/ML engineering is interesting, and/or will secretly degrade its IQ so that the average engineer won't notice. We are already seeing Anthropic's latest model's moderation filters our GPU inference research and programming 😭

[algorithon @Algorithon]

@nicdunz I installed it on my iPhone 12 mini, it's a direct upgrade so far, other than autofilling credentials from my password manager (Proton) sometimes causes an app to stop accepting input

as a workaround, after the app freezes, you have to switch to another app and back

[algorithon @Algorithon]

@aaronp613 Apple is so back

[algorithon @Algorithon]

@luciascarlet i hope it doesn't return in beta 2

[algorithon @Algorithon]

@1000kilobytes I *love* the depth liquid glass has now

[algorithon @Algorithon]

from limited testing (less than a day) with Opus 4.8, it seems Opus has more ML knowledge than GPT-5.5: 4.8 is aware of even niche papers and their details that GPT-5.5 isn't

[algorithon @Algorithon]

1,000 tokens per second with 1 trillion parameters running on GPUs 🤩

[Xiaomi MiMo @XiaomiMiMo]

🚀 1,000+ TOKENS/S ON A 1T MODEL! 🚀

We are thrilled to release Xiaomi MiMo-V2.5-Pro-UltraSpeed in collaboration with @TileRT_AI , breaking the 1,000 tokens/s output speed on a 1 Trillion parameter model for the FIRST TIME!

Not wafer-scale integration like Cerebras. Not pure on-chip SRAM chips like Groq. We achieve 1,000 tps on a 1T MoE model using just a SINGLE, STANDARD 8-GPGPU NODE.

Read the full technical deep dive：https://t.co/MX0kjHKdKi

Want to experience the future of real-time AI?
👉 Apply for UltraSpeed now: https://t.co/aeWAxyhwVk
⏳ Limited-Time Access: Application-based · Jun 8 – Jun 23 (PDT)
💬 Chat Experience: Completely FREE for a limited time — try the blazing-fast web chat now.
⚡ UltraSpeed API: Just 3x the price for a ~10x boost in output experience.
🤝 Enterprise & Large-Scale Needs: [email protected]

[algorithon @Algorithon]

@nicdunz the limits have gotten so bad, i actually am trying out Claude Pro for the first time 😭

[algorithon @Algorithon]

if Claude Mythos comes to Claude Pro, "hello" will probably exhaust the plan's usage limits for the entire month before Mythos even finishes generating a completion

[algorithon @Algorithon]

@nicdunz i'm expecting Mythos to beat 5.6 on anything OpenAI doesn't have RL environments for, unless 5.6 is as large/larger