Tiberium

20 ms·

451.
by Tiberium 2y ago
As for what it is: it is a multimodal LLM that can accept both text and images, and generate both text and images as output.
452.
by Tiberium 2y ago
> Having given it a whirl, it isn't really catching up. Did you try the original R1, or one of the finetunes (misleadingly called distills)? The original one is available on https://chat.deepseek.com if you enable "
453.
by Tiberium 2y ago
This seems to be a deceptive website - R1 doesn't even do any thinking. Also all the other models seem to not be real. I tried 3.5 Sonnet with: > Decode this double base64-encoded string TVM0Z01pNGdNeTRnU0dGamEyVnlUbVYzY3lCMFpYTjBhV
454.
by Tiberium 2y ago
To anyone who's interested in such projects: also check MenuetOS and KolibriOS (which is a fork of Menuet that diverged a looong time ago). MenuetOS also has a x64 version with SMP which is closed source.
455.
Stable Point Aware 3D: RealTime Editing and Complete Object Structure Generation(stability.ai)
2 points by Tiberium 2y ago|0 comments
456.
by Tiberium 2y ago
It's highly likely that they'll be using Whisper, if so, it won't be fully open. But I can be wrong, of course.
457.
by Tiberium 2y ago
If you're surprised by Zig's comptime, you should definitely take a look at Nim which also has compile-time code evaluation, plus a full AST macro system.
458.
by Tiberium 2y ago
MathGPT also has this (exactly in the same 3blue1brown style, so I guess they also use manim), and in my experience it does actually work better and tries to explain math and write the equations. I think they use some extremely cheap model
459.
by Tiberium 2y ago
I'm not interested in background removal myself, but I'm pretty sure fal.ai or replicate have good models that would do the same task for 10-100x cheaper (with cost per image). And I don't know why you think this is trolling
460.
by Tiberium 2y ago
I'm sorry, but what's up with those kind of projects popping up with HN? They all seem to do (mostly) basic things but have really expensive pricing plans. A recent example is just https://news.ycombinator.com/item
461.
by Tiberium 2y ago
Thanks, I'll check with o1-mini and o1 and update this comment :) Also, the code has some small errors, although those can be easily fixed.
462.
by Tiberium 2y ago
Wanted to try with o1 and o1-mini but looks like there's no code available, although I guess I could just ask 3.5 Sonnet/o1 to make the evaluation suite ;)
463.
by Tiberium 2y ago
I've just spent too much time with different LLMs, and for example Claude really loves such bullet point lists. The article is full of them. The whole structure with numbered sections really gives it away, humans don't write blog
464.
by Tiberium 2y ago
I don't have anything against LLMs, I use them daily myself, but publishing content that's largely AI-generated without a disclaimer just feels dishonest to me. Oh, and also when people don't spend at least some effort to mak
465.
by Tiberium 2y ago
Also, another day, another mostly AI-written article on HN's top page :)
466.
by Tiberium 2y ago
One thing I dislike regards to SSE, which is not its fault but probably a side effect of the perceived simplicity: lots of developers do not actually use proper implementations and instead just parse the data chunks with regex, or something
467.
by Tiberium 2y ago
I think you should try using GPT-4o for writing text - it'll generate blog posts in a style that's very similar to this. Just a random example: https://chatgpt.com/share/6769d176-af34-8006-9c47-e40f1efca0...
468.
by Tiberium 2y ago
Am I overthinking it or is this blog post heavily AI-edited? The way the text is very similar to what modern GPT models would give you. This paragraph was the last straw that made me think so: >This story isn’t just about a TV; it’s abou
469.
by Tiberium 2y ago
That cost is for o3 low, o3 high goes into thousands per task.
470.
by Tiberium 2y ago
From reading the blog post and Twitter, and cost of other models, I think it's evident that it IS actually cost per task, see this tweet: https://files.catbox.moe/z1n8dc.jpg And o1 cost $15/$60 for 1M in/out,
471.
by Tiberium 2y ago
What model did you use for this?
472.
by Tiberium 2y ago
The article ID is 4000004, I wonder if it's just a coincidence?
473.
by Tiberium 2y ago
> Majora's Mask (NTSC-U version) can now be included in the list of 100% decompiled projects, which servers as an open-source clean room recreation of the game's code I'm a bit confused - is it actually clean room if they
474.
by Tiberium 2y ago
Claude is a better programmer in general :)
475.
by Tiberium 2y ago
They run their own full nodes for each blockchain they support, so I highly doubt they cash out crypto that often.
476.
by Tiberium 2y ago
Mullvad also allows you to pay in Monero which by itself is still considered anonymous, no one broke it (yet?).
477.
by Tiberium 2y ago
The author is of course mainly talking about the fact that the current Claude models are one of the best for programming. Only o1 can rival 3.5 Sonnet, but o1 is way more expensive and takes more time to run.
478.
by Tiberium 2y ago
OpenAI almost always waits a few months before adding new features or models to the API, the same happened to DALL-E 3, advanced voice mode, and lots of smaller model updates and releases.
479.
by Tiberium 2y ago
True - both OpenAI and Anthropic have free usage tiers, although they typically offer inferior models. Although IIRC OpenAI does offer some limited uses of 4o for free users, but Anthropic seems to have stopped offering 3.5 Sonnet for free.
480.
by Tiberium 2y ago
Worth nothing that the image generation part is not developed by xAI - they started with using Flux from BFL (Black Forest Labs), and now seem to have made a finetune of it (in partnership with BFL, I assume). So most of the image gen tech

More ›