nl

12 ms·

241.
by nl 3mo ago
Another Australian here too. Yes, contacting your MP and senators can be very useful, including for federal stuff. It's harder to actually get meetings with Federal members (they spend a lot of time in Canberra) but still worth trying.
242.
by nl 3mo ago
No? We do GDPR-compliant reporting by using differential privacy to provably remove PII.
243.
by nl 3mo ago
Related: https://news.ycombinator.com/item?id=48517377 It's too bad this has become political. I do differential privacy work for GDPR compliance and it's an interesting technology.
244.
by nl 3mo ago
> variabilities to the chaotic circumstances of real world (“general”) problem solving. All forms of intelligence relate to the reduction of uncertainty. Lossy compression is a form of generalization which handles this exact thing.
245.
by nl 3mo ago
You are going to need to explain yourself in more details. > Compression is not intelligence Just saying this doesn't make it so. It's widely accepted that compression and intelligence have a close relationship. I think this su
246.
by nl 3mo ago
> intelligence negotiates probability (allowing multiple divergent outcomes) while compression requires an idempotent symbolic translation. What does this mean? Lossy, non-deterministic compression is a thing. Does that meet the "a
247.
by nl 3mo ago
Yeah, broadly agree. See my comment on the other story: https://news.ycombinator.com/item?id=48742711
248.
by nl 3mo ago
It isn't distillation that gave GLM 5.2 it's jump in performance. To quote Pat Toulme: There’s a big misconception about how GLM 5.2 was trained. Yes, they distilled Claude and GPT 5.5 — but distillation is not how they matched O
249.
by nl 3mo ago
Opus is still significantly better than open weight models. GLM 5.2 comes close on agentic tasks, but doesn't code as well. Kimi 2.6 and Deepseek v4 Pro write great code but lose track when doing agentic workflows. They were better tha
250.
by nl 3mo ago
Opus isn't nothing! I think the 5x subscription is here to stay - I'd bet they make money on that from lots of people not using it. The 20x is already unavailable in Teams plans.
251.
by nl 3mo ago
This post has been marked as a dupe, but it provides a lot more details than the other announcements of Fable's re-enablement provide: > The export control directive on June 12 came after the government became aware of a report in w
252.
by nl 3mo ago
> In contrast, I've never seen a large model turn bad instructions (instructions that would cause a human to think before starting) into a result I liked I think the distinction is here. I expect my agent to build from product level
253.
by nl 3mo ago
I think this is oversimplifying things. There are many different factions within the administration. Sacks was part of the "deregulate the tech sector" faction, which on this issue is aligned with the "beating China overrid
254.
by nl 3mo ago
Jensen said the Huawei Ascend 950 is roughly comparable to the NVidia H200[1] The H200 was released Nov 2024. Even allowing for Jensen exaggerating the risk there is no way China is 7-10 years behind. Looking at manufacturing process nodes,
255.
by nl 3mo ago
I think you are missing some context here. One of the contributing factors that led to this control in the first place was that the commerce department couldn't get Dario on a call immediately: "Then White House started reaching o
256.
by nl 3mo ago
"Fallback" is only for LLM-training related requests (ie, ones that would compete with Anthropic (!)) For cyber and bio related requests it just refuses.
257.
by nl 3mo ago
I got the TecMojo 10" 6U rack[1] from Amazon for less than $100. It's great, and even came with 2 HTML USB adapter boards (I think for some variety of R-PIs?) Getting something with actual, standard rack-mounting points is worth i
258.
by nl 3mo ago
I love these kind of things - we forget so quickly how much these have improved.
259.
by nl 3mo ago
> This model doesn't shy away from drawing upon all its knowledge. It casually refers to alpha premultiplication and fundamental frequencies in the same breath. It is fond of acronyms. Yes - I had Fable tackle some long-standing bug
260.
by nl 3mo ago
In practice I don't think any harness (happy to be corrected here!) uses the lesser capability models for writing code. The cost trade-offs are rarely worth it. They are often used for reading code though. To expand on this, while the
261.
by nl 3mo ago
> if your ZK vocabulary is limited to stupid BBS-style selective disclosure and nothing else I don't understand what "BBS-style" means in this context, but selective disclosure is exactly what the requirement is.
262.
by nl 3mo ago
Yes, this is true, but that's not what I'm saying. I'm saying that 64GB+ personal computers are vanishingly rare outside builds that were specifically done with AI in mind. Gamers never saw the need for them, and even in soft
263.
by nl 3mo ago
This is so, so good!
264.
by nl 3mo ago
> you find that the GPU often sits idle, not for lack of work, but because the CPU hasn't told it what to do next yet. This phenomenon is called a GPU bubble. This is true, but I've never heard anyone refer to this as a GPU bub
265.
by nl 3mo ago
For identity away from people that knew you or your relatives you had to rely on physical letters of introduction from trusted, well know people. But usually you were proving that you were entering (or leaving) for a legitimate purpose rath
266.
by nl 3mo ago
This is extrapolating from a single example of something that has worked and the conflating correlation with causation. There are plenty of places with highly educated cheap workforces who work hard. Eastern European culture is almost ident
267.
by nl 3mo ago
Yeah, for me it seems like a if you have to ask you can't run it" type question. In general the TL;DR is that anything above 35B needs hardware you buy basically only to run large LLMs, and if you have that hardware you don&#x27
268.
by nl 3mo ago
To add to this, there are numerous well known examples. For example the Bible defined borders of "The Land of Israel" which are still fought over today: https://www.biblica.com/bible/?osis=niv:Ezekiel%2047:13%
269.
by nl 4mo ago
Disagree with this. RISC Zero is useful for crypto use-cases: Other people need to verify an exact program was run. The identity use case is about connecting sources of trust (document issuers) with consumers of that trust ("this is a
270.
by nl 4mo ago
The link above literally says: > Conspiracy to Conceal Documents (Count 12) and other objects that would implicate Maricela Rueda in the riot and shooting at the Prairieland facility. > Defendants convicted: Sanchez Estrada and Marice

More ›