I can't really intuit how I should think about when the model will be accurate. Is there somewhere to read more about that? I assume customers would just have some tests or talk to you.

It looks like the model takes as input a state (structured text? not sure if multi-modal) and a question (as a "Choice", "Score", or "Noul") with some additional augmentations possible. Then outputs the question's answers as appropriate (e.g. a choice, accompanying probabilities, confidence).

Edit: On the AI primer page, it looks like they do the RLCD on a pre-trained base model?

[0]:https://docs.typesafe.ai/concepts/system-one

Comments URL: https://news.ycombinator.com/item?id=49718261

Points: 1

# Comments: 0

Not sure 'copying' is a clarifying explanation of what is happening though.

"factorization ran at no marginal cost on spare or fragmented compute that couldn’t be used for other purposes"

Interesting use of stranded compute.

Do you find this study to be accurate or inaccurate in any ways?

https://pubmed.ncbi.nlm.nih.gov/38728320/

Pdf seems available here: https://www.semanticscholar.org/paper/Not-Everybody-Has-an-I...

I wonder why the scroll performance of every vibe coded app is the same kind of laggy. I assume they're defaulting to the same frameworks or implementations? Or the verification loops in training environments haven't prioritized frame rates/dropped frames yet.

Is it clunky in that it's a verbalized/languified version of system 2 thinking, and clearly humans do some non-verbal version too?

But I do understand that it's a little hard to characterize the amount of compute they're selling while obfuscating the business model enough that it still works. Demand, costs, supply, capabilities, competition change wildly week to week.

Selling a standardized "token" via the API with gas-like pricing works a lot better for transparency. But I think there's a world where the ambiguous subscription pricing is net beneficial for users and providers. Allowing some degree of price discrimination should benefit everyone in aggregate, assuming there's real competition and not yet another duopoly..

The mirror is missing the next post: "Compared to today, this works out to a 17% reduction in weekly limits on Claude Code"

When I submitted the story maybe 8 hours[1] after the announcement, it was already out of stock. This person says it sold out in 1 second - https://x.com/henry19840301/status/2090652363434647738 - if that's possible.

[0]:https://x.com/M5Stack/status/2091116748686922069

[1]:https://news.ycombinator.com/item?id=49390702

Comments URL: https://news.ycombinator.com/item?id=49459641

Points: 1

# Comments: 0

Some labs will be able to buy access. Not every platform will offer it. There are many more labs than the big ones.

Grok searching across x/twitter is too valuable to ignore or silo.

X is effectively a high signal comment layer for the entire internet + quasi internet (arxiv) + real world.

Same applies to any data silo I suppose, X just seems more straightforward to beachhead.

They're always so detailed I've wondered why he doesn't put a Table of Contents, among other editorial decisions..

For perspective - July 2026: 24.9%, All-time low (june 2023): 22.3%, 2008 Financial Crisis Peak: 34.8%, Covid Peak: 34.2%

The twitter and store page have a video of it. There's also a lite version without NFC/LoRa for $55.

1150mAh sounds similar to a kindle. I'm hopeful large programmable/touchable e-ink displays get mainstream soon..

Their store page seems most clear: https://shop.m5stack.com/products/m5papermono-with-lora-nfc-...

Comments URL: https://news.ycombinator.com/item?id=49390702

Points: 2

# Comments: 1