wongarsu

9 ms·

181.
by wongarsu 3mo ago
Website generates a random nounce, the government offers to sign the json { over_18: true, nounce: [inserted_here] }. That signing is coupled to you being signed in, or scanning your passport via nfc or whatever That significantly reduces t
182.
by wongarsu 3mo ago
I believe I have seen 60x improvements on certain classes of bugs. Just not on development as a whole And I've also seen the other side, where vibe coding on projects that had bad code quality to start with leads to bug riddled messes.
183.
by wongarsu 3mo ago
A small mention in your user-wide Claude.md (or Agents.md or whatever you use) about the desired types of comments goes a long way. The typical "why, not what", clarifying larger-picture stuff, and everything else you find in ever
184.
by wongarsu 3mo ago
I'm not sure there are all that many winners. There are those that lose big time with some places potentially becoming uninhabitable, and those that merely have to spend a fortune on new flood prevention infrastructure, deal with massi
185.
by wongarsu 3mo ago
The top of this German article [1] has a rendering of a planned 1.4GW power, 6GWh storage battery plant, with the decommissioned nuclear plant in the background. The battery energy storage is a bit smaller in area and of course of lot less
186.
by wongarsu 3mo ago
If we assume that the forests will regrow to their former strength then the fires are indeed not all the consequential. The trees release the stored CO2 when they burn, and new plants will take up an equivalent amount of CO2 again when they
187.
by wongarsu 3mo ago
Maybe put some quotes around that first paragraph. Seems like some people downvote you before they get to the part where you reveal you don't think that way
188.
by wongarsu 3mo ago
LoRAs or even full fine-tunes would be much cheaper than that, and with some investment in the right infra could be updated regularly. And at least LoRAs can be swapped in and out cheaply, making them usable in large-scale inference provide
189.
by wongarsu 3mo ago
My suspicion is that this is simply due to the M5 having more memory, and the OS already having most of the file cached. The M2 has more memory pressure and would cache fewer of the SSD reads If that's true, inference speed would be ev
190.
by wongarsu 3mo ago
Most of us use a simpler version of the two-agent solution: Claude's auto mode. One agent consumes documents and creates tool calls, another greenlights or refuses them. However this system is somewhat fragile because it depends on the
191.
by wongarsu 3mo ago
Any model with a good score on this benchmark would have a good claim on superhuman abilities. Humans are pretty terrible at being thrown a long policy document and being expected to follow it And while we shouldn't anthropomorphize th
192.
by wongarsu 3mo ago
People understand that. They just don't know how to implement that with LLMs In the GPT-2 era LLMs were just data. Instructions did not exist, and if you added them to your data they would not be followed. Then around 2022 we figured o
193.
by wongarsu 3mo ago
It's not obvious if your core statement is "multiple keyboards is trivial". Assuming that multiple mice obviously means multiple cursors while in the same comment assuming that multiple keyboards obviously does not imply
194.
by wongarsu 3mo ago
Is frontier scale larger than this? Kimi K3 seems to benchmark in the same range as Opus and Fable. I would have expected they are all in the 2-4T range, with quality of the training and architecture differences as the major differentiators
195.
by wongarsu 3mo ago
And don't forget the effect on employee morale, retention and hiring. They have now made the international news cycle for firing an employee for sharing after being encouraged to share, and doing a party trick after being encouraged by
196.
by wongarsu 3mo ago
If you have the choice between AI models created by one evil or the other, it does feel relevant which you consider to be the lesser evil
197.
by wongarsu 3mo ago
The "child pornography" thing was about image generation, which people for better or worse have very different moral standards for than for text generation. There is very little pushback against grok being willing to write explici
198.
by wongarsu 3mo ago
The most valuable element is probably in forming positive interpersonal relationships where none existed previously. Whether that's people that work remotely and rarely see each other in person, or people who work in the same building
199.
by wongarsu 3mo ago
Worth checking first whether your phone already has this natively. My one year old Android by Honor has it built in and exposed in the quick settings (sliding the right down). I know a couple others also have it, but not all of them as disc
200.
by wongarsu 3mo ago
Honor phones have this in the quick settings (the thing that opens when you slide the right side of the top bar down), labeled "Motion Sickness". Felt well-explained and discoverable. But the beauty and pain of Android is every br
201.
by wongarsu 3mo ago
Take the smallest number corresponding to a physically meaningful distance in meters, divide it by two, and that number is still a physical meaningful distance if you switch the unit to decameters or kilometers. Any rational number has some
202.
by wongarsu 3mo ago
Not GP, but my educated guess is that they are running a system with between 4 and 6 MI325 or MI355x or similar AMD GPUs. With the 50k tps as the total figure for all parallel requests. Those cards have a lot of memory for their price, allo
203.
by wongarsu 3mo ago
Where do I sign up to get 200k/yr to keep one rack running? Sounds like an incredibly chill job
204.
by wongarsu 3mo ago
A battery would solve it, but it can be cheaper to have additional panels oriented east and west than to add the equivalent battery capacity
205.
by wongarsu 3mo ago
Exactly. "Is it clean and tidy right now" is a decent proxy for "Has this been properly taken care of in the past" Obviously it's not infallible. But even if the owner only cleaned their stuff now that they are sell
206.
by wongarsu 3mo ago
I've noticed the same pattern even in GLM5.2. It has always been a thing, but it seems to be getting worse in recent models It does feel like the kind of thing beam search would fix. The LLM starts the sentence with a claim like "
207.
by wongarsu 3mo ago
As a citizen, when I have an incomplete understanding of what I'm allowed and not allowed to do I have every incentive to stay on the side of things I'm sure I am allowed to do. Just threatening to do things that you are not allow
208.
by wongarsu 3mo ago
Unlike Tesla, SpaceX is still a visionary and innovative company. Their current cash problems are precisely because they are so heavily invested in so many innovations that have yet to pan out
209.
by wongarsu 3mo ago
I'm not sure "honest" is the right word. Trump loves to lie and make stuff up. It's just usually blatantly obvious, while Trump is also seemingly unable to keep anything secret. Maybe "transparent" is the right
210.
by wongarsu 3mo ago
FTA: 'the firm reported in early June: “Between parts in stock for new builds, and company-owned Reapers with some number of flight hours on them, there are less than 10 total ‘new’ MQ-9As available to any customers anywhere in the wor

More ›