Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
edg5000
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
4 ms
·
31.
▲
by
edg5000
1mo ago
Agree on the false statements on Opus 5. I tested this, 4.8 also got the answer wrong but 4.7 got it right. And so did Sol and Fable.
32.
▲
by
edg5000
1mo ago
Cache reads dominate in modern workflows (coding CLIs and modern web clients such as ChatGPT Work and Claude Cowork (web)).
33.
▲
Sony Music, Warner sue Anthropic for intellectual property theft
(techcrunch.com)
2 points
by
edg5000
1mo ago
|
0 comments
34.
▲
by
edg5000
1mo ago
I spent a long time with custom harness design and found the terminology to be a key part of the work, since the concepts are new. I dropped the term "agent" altogether in favour of "thread" for that very reason.
35.
▲
by
edg5000
1mo ago
> The cost of online services is also largely a result of the cost of training OpenRouter prices are somewhat simmilar to Antrhopic/OpenAI API prices. So I conclude that the hardware plus operating margin alone can genuinely produce
36.
▲
by
edg5000
1mo ago
Sol is 100 EUR/m and I prefer it over Fable, and you get practically unlimited usage unless you go really overboard with it. So I don't get the cost argument. Would be better to own hardware, but with the shortages that's not
37.
▲
by
edg5000
1mo ago
13B is a lot for an online platform that hosts weights. What's the special secret sauce that makes this so valuable? The fact that it's the go-to place for public weights?
38.
▲
by
edg5000
1mo ago
9 to 5 isn't slaving. Even in a jungle, you'd be scavenging and hunting all day. See the birds? They're constantly busy. You want to sit around all day? You'd be miserable. A strange perspective to argue that because pas
39.
▲
by
edg5000
1mo ago
What I do notice is that when it doesn't have the info it needs, or when I don't tell it that I'll handle part of the process, it will indeed do everything to get the job done. Also saw this but find this a great quality, bec
40.
▲
by
edg5000
1mo ago
Sol is basically at the same level, but with better usage. You're not dependent on Fable. You could switch the next minute and hardly experience quality differences.
41.
▲
by
edg5000
1mo ago
In the Netherlands we rely on beach grass, a type of grass that can root well in loose sand, to keep dunes together. This forms the main sea barrier. The grass has historically been replanted at least partially to maintain the protection. T
42.
▲
by
edg5000
2mo ago
The article proves your point. It's eventually proceeded once there was the right history. If he already had the exploit and wanted the model to write an implementation, faking the history could have tripped the model over the edge. So
43.
▲
by
edg5000
2mo ago
It's surprisingly tiny when you exclude the data
44.
▲
by
edg5000
2mo ago
Is anybody in here both engineer and biologist? How far away are we from the complexity of the human body? Pretty far I guess. Although we're producing features at molecular size now with lithography. I choose engineering because I fe
45.
▲
by
edg5000
2mo ago
I'm not happy about it, but API prices seem actually reasonable if you compare it to free-market pure inference providers. They need to buy the same power, same hardware as the big labs without needing any R&D. This means that the
46.
▲
by
edg5000
2mo ago
Initially I was planning heavily around context limits, but I've learned to just ignore it completely. Compaction is seamless for me. If details are lost in compaction, the model just re-reads what's needed. My conclusion is that
47.
▲
by
edg5000
2mo ago
FYI I run it consistently in xhigh regardless of difficulty of the task at hand. I remember high being very fast, but I'd rather wait a bit more and get better output. AIs are insanely fast compared to me anyway, even on xhigh. Consume
48.
▲
by
edg5000
2mo ago
Haha, I remember that rumour being told to me by classmates back in the day. Great example.
49.
▲
by
edg5000
2mo ago
Am I reading this right, and was only the file header encrypted? Also, wasn't ID the company that did not use serials or other DRM? Maybe that explains their naivite, maybe they underestimated how easily games get cracked.
50.
▲
by
edg5000
2mo ago
If OpenRouter prices aren't competitive, the providers would need a cartel to sustain that. It's a commodity market.
51.
▲
by
edg5000
2mo ago
Is this one of those proxies that route to a pool of coding plans? That's the only way to get lower rates than the official Anthropic/OpenAI APIs.
52.
▲
by
edg5000
2mo ago
That's crazy, considering the massive size difference. But the small Qwen models are known for punching above their weight.
53.
▲
by
edg5000
2mo ago
(effort: high) "can the pi 4 use the usb-c port as powered host port when the board is powered via gpio?"
54.
▲
by
edg5000
2mo ago
See my other comment. I have evidence of regression after 4.6. I stopped using Claude altogether.
55.
▲
by
edg5000
2mo ago
We can use OpenRouter pricing to get an idea about what competitive inference pricing is like without R&D or other costs, and indeed we'd be screwed if we had to pay those rates. We'd go from 100-200 USD to 2000-4000 USD/
56.
▲
by
edg5000
2mo ago
You're right. I just re-checked. 4.8 and 5 gave a blatantly wrong answer to a simple question, 4.6, Quen, GLM, Sol gave the right answer. They messed up somehow, not sure what they did.
57.
▲
by
edg5000
2mo ago
Opus 5 as well as 4.8 both gave me a blatantly wrong answer to a simple question, so I dropped them completely. Sol, Qwen and GLM all had the right answer; I only use Sol now. 4.6 had the right answer (I checked with 100% matching prompt),
58.
▲
by
edg5000
2mo ago
From what I hear it's misconfigured bots that can cause issues. This could happen for dynamic content, where the bot thinks there are a lot of unique pages and it just keeps scraping with slightly different query params for example, so
59.
▲
by
edg5000
2mo ago
Arguably no one was really prepared for AI, and NVidia kind of happended to coincidentally have suitable hardware. I have the feeling that something simpler and more efficient is possible when designed for the ground up purely for AI. But I
60.
▲
by
edg5000
2mo ago
Ah, I hadn't thought of that. To what degree have we seen this already? What would you consider as the biggest jump in intelligence per weight?
More ›