Viewing profile — dannyw

Summarised thinking is still sent over the wire via API at least if you request it, inc for Opus 5, inc for Fable 5.

Yeah, the over-cautious filters continues to be a massive pain. There also seems to be bifurcation on consumer (and hence token subsidised) and enterprise (hence API) when it comes…

This is a really cool project. That said, as genuine constructive feedback, I think it would be better if a bit more time was spent on improving the writing and prose of it. I use …

Did you not get all the Codex resets (including banked resets), the removal of the 5hr limit, or even the fact that you can use your Codex sub for API purposes and it's allowed und…

You should try Luna, or look at one of the many independent benchmarks. Even in non-thinking, it's not remotely in the same class as -nano (luna is much smarter).

Oh absolutely! The LLM equivalent of "death and taxes" is prefill and decode; and that holds true irrespective of proprietary inference optimizations. Less verbose output = less co…

Like most companies and releases, OpenAI doesn't flip the switch for everyone at once. I believe Enterprise seats are also generally on a ~2 week delay from consumer.

A lot of users are very latency sensitive. The median ChatGPT.com user is not like the HN community; many of my friends use ChatGPT like it’s the Omnibox in chrome. I’ve literally …

Yes, in the same way a basic, dry hotdog at $5 is bad, but I could love it in $0.50. Pricing absolutely matters.

I did appreciate the article and it's not AI slop by any means, but did anyone else notice how the language and grammar felt very LLM-written? Or at least edited from a LLM draft? …

Logistics and labor. People under-estimate the cost of logistics and transport. Every inch of travel for any input, component, or the final product has to be paid; and even a woode…

The word and use of "engineering" predated what you describe as "real engineering work".

You're getting a free service with real COGS, and the COGS are increasing. 2 weeks notice is entirely reasonable IMO. If I was a paying customer, yes I'd be upset. But... it's lite…

There are architectural solutions. You could duplicate your tokeniser/vocab for example; and have two classes of input: trusted (e.g. system prompts) and untrusted. The exact same …

They did publicly mention some partners have had preview access. Maybe we’ll find out one day.

The MacBooks certainly got a lot better and more usable. Anyone remember the touchpad MBP with no physical escape key and the butterfly keyboard? (Respect to many of Ive’s great le…

There's plenty of sites/APIs providing (actually) uncensored Seedance / Seedream models from ByteDance, which I find surprising too. Yes, they're real Seedance, no refusals.

Very cool project! Sounds like it was a fun challenge :) I wonder when we'll start seeing clusters of ESP32-S3s... not sure how interconnects would go though, but I guess the inter…

Leveraged funds can be an excellent tool for portfolio construction, for example, products like 100% stocks + 100% bonds (so -100% cash; internally borrowed in the ETF), e.g. RSSB.…

I thought OP was referring to AI platforms. For example, by positioning what you’re doing as an accessibility tool and using the right words, you can get the latest models to write…

The idea of companies like OpenAI, Anthropic, Meta, but also NVIDIA (so probably everyone) paying bitcoin (or having conversations about such at least) to 'fugitive sites' like Ann…

I wonder if OpenAI could use a novel defence of “we’re just using it to train AI for better hardware intelligence. Fair use.” It seems to have played out decently in courts so far.…

OpenAI always has the option of transforming themselves to a 'national security corporation' as I'd like to call it. I mean, just look at Kim Dotcom or Aaron Schwartz.

Because Apple has this unorthodox policy of using iCloud for work files, and recommends employees use their personal Apple IDs. I'm not kidding.

Semantic HTML and CSS for text is infinitely better than screenshots for accessibility, as well as cheaper to process for LLMs/agents reading it. For the first reason alone, it's w…