dannyw

11 ms·

211.
by dannyw 2mo ago
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 Anna's Archive for a full download of their ille
212.
by dannyw 2mo ago
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.
213.
by dannyw 2mo ago
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.
214.
by dannyw 2mo ago
Because Apple has this unorthodox policy of using iCloud for work files, and recommends employees use their personal Apple IDs. I'm not kidding.
215.
by dannyw 2mo ago
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 worth it when you can.
216.
by dannyw 2mo ago
That's not just weird, that's completely ridiculous for the most valuable public US company. And it gets more ridiculous when you realise Apple makes the software, the hardware, the OS, and the cloud services. I... really question
217.
by dannyw 2mo ago
According to more than a dozen former Apple employees, apparently not Apple though... despite the fact that they make iPhones.
218.
by dannyw 2mo ago
Apparently Apple encourages their employees to use their personal iCloud accounts for company information, because their own OS doesn't support one iPhone, multiple iCloud identities. Source: https://appleinsider.com/ar
219.
by dannyw 2mo ago
Yay for more cherry-picked information. A substantial amount of evidence in Apple's complaint isn't addressed, but what's shown here throws a decent amount of shade and doubt. I have my popcorn ready for discovery. I hope it
220.
by dannyw 3mo ago
MoE has nothing to do with multimodality. MoE is a concept proposed in 1991, before the deep learning era (which is before what I call the transformers era). You can think of it like sharing. Contrary to popular belief; 'experts'
221.
by dannyw 3mo ago
The for-profit media don’t release photos under freely available licenses. So we’re looking at those images being under copyright for ~100 years. Licenses do matter, just like open source.
222.
by dannyw 3mo ago
Why do you think it's pretty clear? Anthropic models are excellent at working with Blender APIs, other industry standard 3D modelling programs and tasks, and game libraries that have nothing to do with Three.js. Three.js is a quite pop
223.
by dannyw 3mo ago
Some people here think Wikimedia Foundation's mission is to keep Wikipedia up. That's a subset of their mission. I attribute nearly all of the blame to WMF, because their donation ads are quite deceptive. The actual, public missio
224.
by dannyw 3mo ago
I'm generally critical of WMF and think (1) they waste significant amount of resources on mis-managed projects, and (2) are not very honest or transparent in donation communications. But to play devils advocate, a lot of this depends o
225.
by dannyw 3mo ago
Those absolutely are miracle numbers and don’t seem remotely accurate. What does the wall report? I remember people were saying this IDLES at like 80w wall, I’m sure it’s been optimised, but 35w at load is super unlikely
226.
by dannyw 3mo ago
But a LLM literally does assign a token from a fixed set of tokens (its vocab) to an input, rinse and repeat, until the stop token. Classifiers have been giving logits since decades ago.
227.
by dannyw 3mo ago
A lot of world-class talent do not want to pay 50-60% taxes, so they move somewhere else. Not everyone, I'm just saying, it was a genuine consideration for me.
228.
by dannyw 3mo ago
Only for Anthropic. OpenAI officially allows you to use your subscription in non-Codex harnesses / API-like use.
229.
by dannyw 3mo ago
They dropped the -preview ID.
230.
by dannyw 3mo ago
Actually, saving $1 on something you would pay anyway is more valuable than earning $1 of taxable income. I think that's just semantics -- unless you genuinely keep racking up non-redeemable energy bill credits for feed-in.
231.
by dannyw 3mo ago
Oh, LLMs are incredible at solving captchas when allowed to. I mean, just download an abliterated version of Gemma4 E4B even, it'll solve pretty much any captcha.
232.
by dannyw 3mo ago
Qwen/Alibaba have stopped doing open weights releases for a while. No grudge or anything, I'm certainly not going to look at a gift horse in the mouth, but both DeepSeek and Moonshot have been very consistent with open weights as
233.
by dannyw 3mo ago
Usable local AI has been here for a while, esp on say a 5090. You can’t treat Qwen3.6 like its fable, but if you prompt precisely and specifically it’s a great executor. I actually found it refreshing to use more of my brain for once, and a
234.
by dannyw 3mo ago
Deepseek and moonshot are the only two providers I consent to training for.
235.
by dannyw 3mo ago
General light/defined tasks, e.g. go fetch a list of rows from <X> and then run script <Y> against each. Yes, I can just do it myself, but even at API prices, I'd rather have the LLM do it.
236.
by dannyw 3mo ago
Smaller models are also excellent executors/operators. When given a solid plan, including e.g. what files/functions to change and how, what to test, even a recent 27-30B open weight model will usually do an excellent job.
237.
by dannyw 3mo ago
Was that Codex/subsidised-usage or API? I do get overloaded in Codex-account from time to time, but API is rock solid. They obviously load shed a bit of Codex-sub during peak times, and for the amount of tokens you get for a sub, I do
238.
by dannyw 3mo ago
You don't see real reasoning tokens with the Chat Completions API; and 'store: false' with Responses really isn't that different to Completions. FWIW, if you have some tokens to spend, you might want to test Responses vs
239.
by dannyw 3mo ago
I'm in complete agreement that things like encrypted sub-agent messages, etc, are terrible, especially because it is fundamentally irresponsible to have an agent running on your machine without the slightest hint of what the prompt is.
240.
by dannyw 3mo ago
Have you seen https://arxiv.org/html/2507.14805v1 ? It's genuinely a novel space. I found the write-up to be interesting, although I do agree 486k tokens is not sufficient for evaluating LLMs (even with greedy dec

More ›