dannyw

11 ms·

181.
by dannyw 2mo ago
One window that can be shortened is working with software ecosystem and upstream partners; think day 0 on together, fireworks, Unsloth, etc. That obviously happens from partners getting embargoed weights early.
182.
by dannyw 2mo ago
I personally don’t like, and wouldn’t work for Meta; but it’s an Apache 2.0 model. I’m liking it, and I don’t see a personal moral contradiction here. Do you use React for frontend for example? I also wish this HN post is a bit more focused
183.
by dannyw 2mo ago
For the same price as a DGX Spark here (A$8499) I can buy roughly 544GB of DDR5-5200MHz from retail; which on a quad channel platform would deliver ~160gb/s real world; and ~320gb/s with octa channels (Xeon, Threadripper Pro). If
184.
by dannyw 2mo ago
The XML tags are similar to <antml:xxx>, which is obviously Anthropic ML (or ANTrophic xML). I think it’s likely 3; meta in reverse. While tokenisers and preprocessing can catch it, you want your special tokens to be unique and not pr
185.
by dannyw 2mo ago
You can use another harness like Pi, OpenCode, etc and build your own auto reviewer if you’d like (or adapt the open source Codex one). If you have an openai subscription you are explicitly allowed to use your subsidised tokens / usage
186.
by dannyw 2mo ago
Plus extremely easy to ablate, abliterate, apply steering vectors, etc. DeepSeek V4 Flash is good enough to do a 1-shot steering vector sweep on itself. Weights might seem like random numbers, but having weights gives you amazing, control,
187.
by dannyw 2mo ago
That's so subjective. The idea isn't new, many have proposed twice-annually earnings reports; for example the LTSE (Long-Term Stock Exchange) has been trying to make it happen for a while. If you have worked at a public or large c
188.
by dannyw 2mo ago
Maybe by design? 1. Metrics show auto or YOLO mode users have higher retention 2. Company intentionally doesn’t invest in manual mode
189.
by dannyw 2mo ago
I’d rather use another open source solution that doesn’t require a signup, and less likely to get rugpulled. There is no reason to require a login for creating local mini sandboxes. If you’re on Apple, native solutions like “container-machi
190.
by dannyw 2mo ago
Thank god. The actual patent (roughly; I researched it myself, but a few years ago so memory fuzzy) involves compressing frames/video before debayering which makes the streams more easily compressible; instead of after debayering, and
191.
by dannyw 2mo ago
Summarised thinking is still sent over the wire via API at least if you request it, inc for Opus 5, inc for Fable 5.
192.
by dannyw 2mo ago
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 to 'safeguards'. The exact same prompts hit safeg
193.
by dannyw 2mo ago
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 LLMs to write too, and give it examples of my writing for st
194.
by dannyw 2mo ago
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 under their terms? I dunno, I've honestly never been
195.
by dannyw 2mo ago
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).
196.
by dannyw 2mo ago
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 context & less token gen.
197.
by dannyw 2mo ago
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.
198.
by dannyw 2mo ago
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 seen people ask ChatGPT for a link to Gmail. Is it a dark pa
199.
by dannyw 2mo ago
Yes, in the same way a basic, dry hotdog at $5 is bad, but I could love it in $0.50. Pricing absolutely matters.
200.
by dannyw 2mo ago
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? I used to really enjoy Chips and Cheese's writing,
201.
by dannyw 2mo ago
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 wooden fencepost typically moves between at least 5 different loc
202.
by dannyw 2mo ago
The word and use of "engineering" predated what you describe as "real engineering work".
203.
by dannyw 2mo ago
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 literally free... and it's still free. Compa
204.
by dannyw 2mo ago
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 phrase can tokenize differently depending on if it&#x27
205.
by dannyw 2mo ago
They did publicly mention some partners have had preview access. Maybe we’ll find out one day.
206.
by dannyw 2mo ago
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 legacy though)
207.
by dannyw 2mo ago
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.
208.
by dannyw 2mo ago
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 interconnect wouldn't be the bottleneck anyway.
209.
by dannyw 2mo ago
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. And just because it's available doesn't mean it s
210.
by dannyw 2mo ago
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 incredibly powerful malware without safeguards kicking in.

More ›