Hacker News: mkotlikov

New comment by mkotlikov in "U.S. government will decide who gets to use GPT-5.6"

mkotlikov — Fri, 26 Jun 2026 21:59:39 +0000

US will ban Chinese models and try to get their allies to do the same. Just like they did with Huawei. Alternatively, they'll put up legal roadblocks that open models are unlikely to jump over due to costs or other reasons.

Otherwise they're putting US frontier labs at a huge disadvantage by preventing them from recouping costs on their biggest models.

How much more will OpenAI and Anthropic models cost when they're the only AI you can legally use?

New comment by mkotlikov in "Claude Fable 5 silently degrades its own performance on frontier AI work"

mkotlikov — Wed, 10 Jun 2026 09:29:21 +0000

Anthropic's new Fable 5 system card describes a safeguard for "frontier LLM development" requests. Unlike the cyber/bio/distillation safeguards, this one doesn't refuse or fall back to another model. It silently reduces the model's effectiveness via prompt modification, steering vectors, or PEFT, with no notification.

This bugs me for 4 reasons...

1) The named examples (pretraining pipelines, distributed training infra, ML accelerator design) are just normal ML.

2) A false positive on a silent intervention is undetectable by design: you can't distinguish a degraded answer from a hard problem or your own bug.

3) The 0.03% figure is self-reported against a private benchmark with no way for anyone outside to audit it.

4) A degraded answer is still charged at full price.

So...Anthropic is now purposely releasing a dishonest model? What am I missing? Tell me how I'm stupid and/or wrong!

Claude Fable 5 silently degrades its own performance on frontier AI work

mkotlikov — Wed, 10 Jun 2026 09:29:21 +0000

Article URL: https://mkotlikov.substack.com/p/fable-5-considered-harmful

Comments URL: https://news.ycombinator.com/item?id=48473737

Points: 1

# Comments: 1

AI Is Starving for PDFs

mkotlikov — Sun, 17 May 2026 21:20:54 +0000

Article URL: https://mkotlikov.substack.com/p/your-ai-is-starving-for-pdfs

Comments URL: https://news.ycombinator.com/item?id=48173305

Points: 3

# Comments: 0

New comment by mkotlikov in "Electromechanical reshaping, an alternative to laser eye surgery"

mkotlikov — Tue, 19 Aug 2025 17:54:59 +0000

I use ortho k, I exclusively sleep in my contact lenses, take them out when I wake up and enjoy 20/20 vision for 24+ hours. I don't have to sleep in them every night to maintain 20/20, but I do anyways.

They are basically retainer contact lenses that reshape your cornea as you sleep. It's what's described here minus the pH change and electric potential.

New comment by mkotlikov in "Benchmarking GPT-5 on 400 real-world code reviews"

mkotlikov — Fri, 08 Aug 2025 08:00:20 +0000

Models tend to prefer output that sounds like their own. If I were to run these benchmarks I would have:

1) Gemini 2.5 Pro rank only non-google models 2) Claude 4.1 Opus rank only non-Anthropic models 3) GPT5-thinking rank only non-OpenAI 4) Then sum up the rankings and sort by the sum.