Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
onion2k
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
6 ms
·
91.
▲
by
onion2k
2mo ago
Wealthy patronage has always steered art, and the CIA has a lot of money.
92.
▲
by
onion2k
2mo ago
I learned HTML when I was 17 in about 1995 and it's certainly taken me on a pretty fun career path. Less technical than LLMs for sure, but 'figure out where the industry is going and move what you're learning to there' i
93.
▲
by
onion2k
2mo ago
Claude has a /goal function that explicitly says it'll carry on working until it's done what you prompted it to do. That's exactly what a 'AI will zero-shot anything' believer is looking for. Behind the scene
94.
▲
by
onion2k
2mo ago
The bigger models are having trouble right now presumably regardless of access method because Claude Code isn't working for them either - https://status.claude.com/ - switching to Sonnet fixes the problem though. As an
95.
▲
by
onion2k
2mo ago
does this failure mean I need to change something about this system? It's safer to say 'does this failure mean I need to change the system?'. Otherwise you're still facing the single cause trap.
96.
▲
by
onion2k
2mo ago
Hackernews, BBC.co.uk (I'm British, I trust it), Reddit, and YouTube. I informed about something , but whether it's actually right is a separate question.
97.
▲
by
onion2k
2mo ago
I've been using Opus 5 to write a fractal renderer in GLSL today, with pretty good results (better than I could do on my own anyway). It definitely can do basic maths.
98.
▲
by
onion2k
2mo ago
Surely the level of stress needed to have this impact is something that needs to be studied. The researchers presumably went to an extreme to ensure a clear signal for a result, but the fact they've shown there is a significant outcome
99.
▲
by
onion2k
2mo ago
The separation in micro services is hard to break though. You need to add an API on one side and a call on the other. In a monolith it's significantly easier to do something that has an unexpected impact somewhere else. For example, ch
100.
▲
by
onion2k
2mo ago
There's a youtuber called Switch Angel who regularly posts live coding sessions of her making trance music with Tidal. Worth a listen and a watch to see what's possible. https://youtube.com/@switch-angel?si=wUTkBU1
101.
▲
by
onion2k
2mo ago
I don't think the author is suggesting we shouldn't build useful things. He's saying to the people who built the military system that they shouldn't be doing that. It's the last step that's the problem, not the
102.
▲
by
onion2k
2mo ago
I'd be interested to overlay, I don't know, customer satisfaction or anything that can show the follow-on effect of all this output. Linear won't have that information. Very few businesses can accurately attribute customer
103.
▲
by
onion2k
2mo ago
This looks like measuring what is easy to do, rather than what really matters. Even that's hard. There aren't enough signals to attribute changes directly to AI, so these apps seem to correlate the signal that the user was int
104.
▲
by
onion2k
2mo ago
That implies you think the output of an LLM is only a rewording of the prompt, and that it doesn't add anything from it's corpus of training data. That's obviously untrue.
105.
▲
by
onion2k
2mo ago
In a strictly libertarian sense, sure. In more liberal sense, because you might be destroying something unique that later generations might actually like to see, the life rule that states "Don't be a dick" probably applies tr
106.
▲
by
onion2k
2mo ago
Practically, no, the distill is great. It's fine to use it. However, if you're having a discussion about access to frontier models, and using Qwen 3.8 as an example of how open weights is a solution, then you should be honest and
107.
▲
by
onion2k
2mo ago
When people talk about Qwen 3.8 being on a par with Fable, they're really talking about Qwen 3.8 Max aka Qwen3.8-2.4T-A95B. That's a 2.4 trillion parameter Mixture of Experts model with 95B active parameters. You need about 400GB
108.
▲
by
onion2k
2mo ago
Tells you a lot about the man, doesn't it. The CEO of Anthropic not using AI extensively would be news.
109.
▲
by
onion2k
2mo ago
I think it's more that it shifts the thought process from "If this is going to take 30 years then we won't bother, because the investment can be spent on things that pay off sooner" to "If we can do this in 3 years
110.
▲
by
onion2k
2mo ago
That's one of the major things the AI community is still completely missing. That isn't true. It's not continuous like in humans, but it's clear that models are using prompts, feedback, etc to improve. They're lea
111.
▲
by
onion2k
2mo ago
Ah, research in the pre-LLM era was quite different. You actually had to know which questions to ask. You get a much better response if you ask the right questions in the right language though. Expertise still has value. I guess you just
112.
▲
by
onion2k
2mo ago
Randomness without clusters probably isn't random.
113.
▲
by
onion2k
2mo ago
Upvoted your post. Muhahaha!
114.
▲
by
onion2k
2mo ago
Illegal? Hell no. Spending VC money to capture a market and drive out the inefficient incumbents is a genuinely good plan. Market disruption is a massive net benefit. I just don't think it'll work in the case of Temu, and anyone w
115.
▲
by
onion2k
2mo ago
You missed the biggest one: * Not Nigel Farage
116.
▲
by
onion2k
2mo ago
If someone says to you "You can buy this widget that costs $10 to make for $5" that doesn't mean they they're selling it for $5. It means there's more going on and that the $5 price take is not the full story. You
117.
▲
by
onion2k
2mo ago
A few years ago I would have laughed at anyone telling me that there is a serious market for ten-dollar drills, two-dollar dresses, and one-dollar pairs of shoes shipped from a warehouse on the other side of the planet. TEMU doesn't
118.
▲
by
onion2k
2mo ago
I have fond memories of Slashdot in 1998. It was the spiritual forerunner to HN, and had upvoting, and in a more complex form as there were categories of votes. I got lots of +5 Funny comments. Hard to imagine, isn't it.
119.
▲
by
onion2k
2mo ago
It's worth flipping the question around to explore how you actually think. So rather than ask "What is the impact of AI?" which narrows your focus to the impact of AI, if you really care about the natural resources you should
120.
▲
by
onion2k
2mo ago
Fertiliser and pesticide runoff from golf courses into water systems is a well known environmental issue.
More ›