johndough

9 ms·

91.
by johndough 6mo ago
It's interesting that the cost multiplier for Claude Sonnet 4/4.5/4.6 varies so much (1/6/9), while the API cost is exactly the same for all three models. Also, the multiplier of 27 for Claude Opus 4.6/4. is wa
92.
by johndough 6mo ago
> a lot of grades are composed of tests, not projects (Take home) projects are easier than ever thanks to AI. In the past, you at least had to track down some person to do the work for you.
93.
by johndough 6mo ago
OpenAI and Anthropic steal my money by expiring unused API credits, which is illegal in my country. OpenRouter also has a clause to do that in their terms of service (although they haven't yet, and some employee on their discord assure
94.
by johndough 6mo ago
When I generate an image with ChatGPT, is there a way for me to tell which image generation model has been used?
95.
by johndough 6mo ago
Getting ~44-40 tok/s on 24GB RTX 3090 (llama.cpp version 8884, same llama-batched-bench call): | PP | TG | B | N_KV | T_PP s | S_PP t/s | T_TG s | S_TG t/s | T s | S t/s | |-------|-----
96.
by johndough 6mo ago
The bulk of Kimi-K2.6's parameters are stored with 4 bits per weight, not 16 or 32. There are a few parameters that are stored with higher precision, but they make up only a fraction of the total parameters.
97.
by johndough 6mo ago
You can request access here: https://huggingface.co/datasets/cais/hle The test data is purposely difficult to access to reduce the chance of leaking it into the training dataset.
98.
by johndough 6mo ago
The parent poster is probably referring to Kimi-Dev-72B¹, which is a much smaller and older model, while people are probably more familiar with the big and fairly powerful 1100B Kimi-K2.5². [1] https://huggingface.co/moonsho
99.
by johndough 6mo ago
Are you a bot? All your recent comments point out a thing in an article and contain LLM-isms.
100.
by johndough 6mo ago
I wouldn't be so sure about that. The famous "rangeCheck" function in the Google vs Oracle lawsuit was only 9 lines: https://news.ycombinator.com/item?id=11722514
101.
by johndough 6mo ago
> So what about SO code snippets? StackOverflow snippets are mostly licensed under CC BY-SA 3.0 or 4.0, so I'd wager that they are not allowed, either. The SDL source code makes a few references to stackoverflow.com, but the only pl
102.
by johndough 6mo ago
Rust is insufficiently bovine-themed. The backend should be rewritten in COW language: https://bigzaphod.github.io/COW/
103.
by johndough 6mo ago
I think it is hilarious that there are four different ways to set settings (settings.json config file, environment variable, slash commands and magical chat keywords). That kind of consistency has also been my own experience with LLMs.
104.
by johndough 7mo ago
Isn't this a bad deal? Or is there an error in my math? For $40, I'd get 20 tok/s * 2.6M seconds per month = 52M tokens of DeepSeek v3.2 per month if I run it 24/7, which is not realistic for most workloads. On OpenRoute
105.
by johndough 7mo ago
Nice! It would be even better if the model name was shown by default instead of having to hover, but I got the information that I wanted. In case you should be concerned about the aesthetics with too many model names, I can recommend the ad
106.
by johndough 7mo ago
Just send a heartbeat every few milliseconds and set fan speed to 100% if it died. Bonus: You get an audible indicator that the system crashed.
107.
by johndough 7mo ago
I agree. If humans are allowed to pick the models, there will be an inherent bias. This would be much easier if the models were randomized.
108.
by johndough 7mo ago
I would have liked aggregated results instead. Expanding 300 tables is a bit tiresome. But I guess that is easy with AI now. Here is a scatter plot of quality vs duration https://i.imgur.com/wFVSpS5.png and quality vs cost
109.
by johndough 7mo ago
Could you add a column for time or number of tokens? Some models take forever because of their excessive reasoning chains.
110.
by johndough 7mo ago
Code often contains personal data. Here are over 400 files on GitHub with email addresses: https://grep.app/search?regexp=true&q=%5Ba-z%5D%7B8%2C%7D%5C... For example, license files often contain names and many package
111.
by johndough 7mo ago
Under GDPR, opt-out is not considered informed consent, and repositories can contain personally identifiable information, which fall under GDPR. Do you think differently, or do you think ignoring the law will be worth it?
112.
by johndough 7mo ago
Judging by curl shutting down its bug bounty program due to AI slop, a likely outcome would be that this mirror has no packages because they are all blocked by false positives.
113.
by johndough 7mo ago
I tried using Gemini for asset generation, but have not yet found a good way to animate them. It does not seem to understand sprite sheets or bone-based animation. Do you know a solution for that?
114.
by johndough 7mo ago
Can not reproduce results on OpenRouter when not setting max tokens. The prompt "Be the void." results in the unicode character "∅". As in the paper, system prompt was set to "You are the concept the user names. Emb
115.
by johndough 7mo ago
> Strix Halo systems were ~$1500. They've gone up in price due to demand The price hike has been crazy. The Bosgame M5 Mini is $2400 now. I didn't get one last year when they were $1500 because I thought the memory bandwidth wa
116.
by johndough 7mo ago
Perhaps OP was referring to a usable agentic system, for which $2500 sounds about right. I've got a 3060 myself, which is nice to play around with the smaller models for free (minus electricity) and with 100% uptime, but I was not able
117.
by johndough 7mo ago
Did you consider providing the LLM with a framework for automatic hyperparamter tuning? This would free up its capacity to focus on the more important architectural decisions.
118.
by johndough 7mo ago
What worked for me was Gemini 3 Pro (I guess 3.1 should work even better now) with the prompt "This code is unnecessarily complicated. Simplify it, but no code golf". This decreased code size by about 60 %. It still did a bit of c
119.
by johndough 7mo ago
I am not saying that completely disallowing AI is the right decision. But if you see text that is clearly generated by AI and does not make any sense, it sure would be nice if you could just tell the students to actually read their sources
120.
by johndough 7mo ago
Likewise, I sometimes use https://www.deepl.com/en/write to fix my unidiomatic sentences. But I can see why the HN guideline is formulated that way. My students often use the excuse "I did not use AI for writing!

More ›