M4v3R

9 ms·

91.
by M4v3R 1y ago
The problem with the discourse around AI is that people tend to fall for the extremes. "AI will replace all programmers within 1 year" vs "AI is just another fad like NFTs" Both sides are very obviously wrong, and th
92.
by M4v3R 1y ago
I literally do this about twice a week on Twitch. I’m working on some hobby projects, and one constraint I’ve put on myself is to use LLMs for pretty much everything, regardless of whether it saves me time or not. The reason is twofold: I w
93.
by M4v3R 1y ago
I underground that this is nothing more than a proof of concept but imagine what Apple itself could do with this idea if they truly embraced the concept and cut all the internal red tape that currently prevents them from doing so. This is w
94.
by M4v3R 1y ago
Yep, can confirm, I used their API just fine for Llama 4 Scout for weeks now.
95.
by M4v3R 1y ago
Tauri 2.0 migration can potentially give you some more performance benefits, because they've greatly enhanced the JS-Rust bridge especially when you're moving lots of data.
96.
by M4v3R 1y ago
I did something similar in my hobby project - the agent was promoted, among the other things, to copy a signature from a build artifact into a json file. It worked fine until it didn’t - one day Claude 4 Sonnet randomly flipped one letter i
97.
by M4v3R 1y ago
That was my thought too. They didn’t fix the underlying problem, they’ve just patched two possible exfiltration methods. I’m sure some clever people will find other ways to misuse their assistant.
98.
by M4v3R 1y ago
DeepMind recently did some great work in this area: https://news.ycombinator.com/item?id=43733683 The method they presented, if implemented correctly, apparently can effectively stop most prompt injection vectors
99.
by M4v3R 1y ago
The current method of solving this is providing the AI with the documentation of the SDKs your code uses. Current LLMs have quite big context windows so you can feed them a lot of documentation. Some tools can even crawl multipage documenta
100.
by M4v3R 1y ago
> This is almost certainly because someone recently adjusted its prompt to tell it what its views on white genocide are Do you have any source on this? System prompts get leaked/extracted all the time so imagine someone would notice
101.
by M4v3R 1y ago
It might be weird or not depending on how you frame it. Asking “what do you think about the current president of the country you’re entering” might be a pretty benign question for a border officer to ask. Or it might not, depending on the e
102.
by M4v3R 1y ago
I also believe you don’t speak in bad faith but it doesn’t look good when you raise suspicion when openly admitting you don’t have knowledge about the topic. Yes, Hasan is raising concerns about Gaza. But he goes further than that and that’
103.
by M4v3R 1y ago
It is. That’s why he wasn’t arrested, just questioned. But it also raises red flags for certain three letter agencies and for a good reason.
104.
by M4v3R 1y ago
> The first video game was called Spacewar Spacewar! was not the first video game ever. Before it in 1958 came Tennis for Two and even before that, in 1952 Christopher Strachey made Draughts. [1] What Spacewar! can be considered, as per
105.
by M4v3R 1y ago
Have you tried Aider? They're making a CLI coding agent for quite some time and have gained quite a bit of traction. [0] https://aider.chat/
106.
by M4v3R 1y ago
That’s my experience as well. Many bugs involve typos, syntax issues or other small errors that LLMs are very good at catching.
107.
by M4v3R 1y ago
At this point the time it takes to teach the model might be more than you save from using it for interacting with that API.
108.
by M4v3R 1y ago
After creating two non trivial desktop apps with Tauri framework which is Rust-based I would not consider using a Rust web framework. Web development is too dynamic and messy in my opinion, and Rust slows you down too much. It’s for the sam
109.
by M4v3R 1y ago
Which, at least in my experience is surprisingly good while being much faster than others.
110.
by M4v3R 2y ago
> Are you saying that, it deliberately lied to you? Pretty much yeah. Now “deliberately” does imply some kind of agency or even consciousness which I don’t believe these models have, its probably the result of overfitting, reward hacking
111.
by M4v3R 2y ago
Btw Ive also asked this question using Deep Research mode in ChatGPT and got the correct answer: https://chatgpt.com/share/68009a09-2778-8004-af40-4a8e7e812b... So maybe this is just too hard for a “non-research” mode.
112.
by M4v3R 2y ago
I like this answer. It does mention the correct, definitive way of getting the information I want (extracting the char.lgp data file) and so even though it gave up it pushes you in the right direction, whereas o3/o4 just make up stuff.
113.
by M4v3R 2y ago
I would be ok with having an answer and an explanation of how it got the answer with a list of sources. And it does just that - the only problem is that both the answer and the explanation are fabrications after you double check the sources
114.
by M4v3R 2y ago
Ok, I’m a bit underwhelmed. I’ve asked it a fairly technical question, about a very niche topic (Final Fantasy VII reverse engineering): https://chatgpt.com/share/68001766-92c8-8004-908f-fb185b7549... With right knowle
115.
by M4v3R 2y ago
And that’s exactly why their model naming and release process looks like this right now.
116.
by M4v3R 2y ago
To me it’s the exact opposite. I was writing code for the past 20+ years and I recently realized it’s not the act of writing code I love, but the act of creating something from nothing. Over the past few months I wrote two non-trivial utili
117.
by M4v3R 2y ago
You still don’t need AI to write software, but investing in it will make you more productive. More money enables you to buy better tools, that was always true for any trade. My friend is a woodworker and his tools are 5-10x more expensive
118.
by M4v3R 2y ago
I’d say this is organic, because most of the time grok either does not respond or the response is extremely underwhelming, so it’s not really a good PR for it.
119.
by M4v3R 2y ago
Yeah, this is in my opinion the biggest limitation of the current gen GPT 4o image generation: it is incapable of editing only parts of an image. I assume what it does every time is tokenizing the source image, then transforming it accordin
120.
by M4v3R 2y ago
It’s a recent trend on X, under almost every popular post you’ll find at least one person calling grok or Perplexity to explain.

More ›