nl

25 ms·

331.
by nl 4mo ago
OVH is acting as a "Deployer", not a "Provider", which have special meaning under the AI Act. There are much less (almost no) disclosure regulations on the deployer. https://ethicalogic.com/articles/
332.
by nl 4mo ago
Qwen 3.5 and above use Gated DeltaNet which alternate attention and SSM layers: https://sebastianraschka.com/llms-from-scratch/ch04/08_delta...
333.
by nl 4mo ago
> what Alibaba spends their time doing? Not most of the time (pre-training takes a long time), but post-training is where most of the value is, yes. Famously it is all that OpenAI did between GPT 4o and GPT 5.3 (or 5.2?) - they didn&#x
334.
by nl 4mo ago
Apple didn't even have an AppStore in mind until people started hacking apps onto it. Remember back in 2007 when Apple first told developers that to develop for the iPhone, they’d need to build WebApps for Safari? Well, that really wa
335.
by nl 4mo ago
it says: Proxy (production) For production, route requests through your own back end with .proxied. The relay at baseURL adds the Claude API credential server-side, so the app ships no key. The headers you provide are sent on every request
336.
by nl 4mo ago
Handy is the most common recommendation https://handy.computer/
337.
by nl 4mo ago
> The analogy falls apart very quickly. Without the training data, your modifications amount to virtually nothing compared to what these "versions" are, and the idea that you can maintain and improve on these models without the
338.
by nl 4mo ago
> so we took the census for centuries before this point, and it was “ok.” It wasn't ok - it's been shown that the data released could individually identify people in releases before the 2010 Census.
339.
Noise infusion banned from statistical products published by Census Bureau(desfontain.es)
899 points by nl 4mo ago|604 comments
340.
by nl 4mo ago
Not sure if you are agreeing with my point or not. But I assure you that many places will be happy to switch up to Fable when it's available and back to Opus when it's not. It's a programming tool, not something that will sen
341.
by nl 4mo ago
VPNs won't work when they do document (passport) verification.
342.
by nl 4mo ago
I'm a strong proponent of Open Source (TM) but I disagree with this take. The weights are the useful artifact here. You can modify them, fine tune them and do what you want with them. Unlike binary software there is nothing limiting th
343.
by nl 4mo ago
When kubernetes was released there were very few people who could run it, and even less that could run it usefully. Right now there a few people who can run a 1T model at home, even less who can run a 5T model and probably single digits who
344.
by nl 4mo ago
> Trump says his team will 'look into' US taking stake in AI companies[1] Yes, there is a gap between "taking a stake" and nationalizing one, but.. [1] https://www.reuters.com/business/trump-says-
345.
by nl 4mo ago
"It is time to go beyond transparency to more serious and binding regulation of AI."[1] Anthropic is calling for regulation. For example they endorsed CA SB-53 that even OpenAI and Google thought was too much: https:/&#x2F
346.
by nl 4mo ago
Sovereign AI is about to get hot. It's difficult to predict this administrations actions, but given it included employees that has to be a huge risk for Google, where Deepmind is based in London. Cohere (Canada) and Mistral (France) ar
347.
by nl 4mo ago
I use both Opus and Fable on tasks that are well beyond "things that would take a human 3 hours" It fails all the time - as in it ends up doing something I want to change. But this doesn't actually matter - if it takes 3 or
348.
by nl 4mo ago
Not yet that I've tried, and I'm pretty systematic about test driving them. I keep https://sql-benchmark.nicklothian.com/#all-data up-to-date with latest releases and try out most that score 24+. GPT 5.5+ or Opus
349.
by nl 4mo ago
> Small and fast experts with a router LLM on top to best use them A router LLM isn't a MoE. A MoE is a type of LLM architecture, not lots of different LLMs. They are fundamentally different concepts and it is a fundamental misund
350.
by nl 4mo ago
Non sequitur
351.
by nl 4mo ago
> Did you manage to setup a discussion with the agent to reveal such assumptions? This is what the plan review is for. Usually it will have something like "modify abc.ts to update the widget number in the wnx collection" and I
352.
by nl 4mo ago
Confirmed by you ! I don't really agree with their point here, but there are plenty of people in the AI community whose views are aligned with Anthropic's. That doesn't make them shills. It's actually important those vi
353.
by nl 4mo ago
I think it's a big mistake to conflate the cyber (and bio) refusals with the LLM development refusals. I can sympathize with the argument for the cyber refusals - especially as a temporary measure - especially if Mythos is available to
354.
by nl 4mo ago
The kinds of detailed (and excellent) plans Opus or Fable can generate on our large code base would take me maybe 1-2 days to work through and they do in 10-20 minutes. Maybe I spent 2-4 hours reviewing it, checking things with colleagues e
355.
by nl 4mo ago
What do you mean? You absolutely can run compression in WASM. For example here is Gzip in WASM: https://github.com/ColinTimBarndt/wasm-gzip
356.
by nl 4mo ago
A word of caution on this. I've tried this too, and was disappointed. Kimi generally benchmarks at "a bit more intelligent than Sonnet Medium" levels[1] and I'd agree broadly with this assessment. If you have adapted you
357.
by nl 4mo ago
This is different to the cyber limitations though. To be precise - it makes the "won't work on frontier machine learning" refusal the same as the "won't work on cyber security" refusal (instead of the way it pr
358.
by nl 4mo ago
I think that's the point the OP is making...
359.
by nl 4mo ago
The new data retention policy is interesting. Seems to apply even to enterprise plans on ZDR. > Finally, we’re making a change to the way we handle business customer data for Fable 5, Mythos 5, and future models with similar or higher ca
360.
by nl 4mo ago
Whether or not consumers wanted it, Apple failed to deliver. And that wasn't because they were listening or anything - they just couldn't deliver. Apple doesn't leak much but there has been coverage of this: https:/&#x2

More ›