101–110 of 415 posts
I see the value in showcasing that LLMs can run locally on laptops — it’s an important milestone, especially given how difficult that was before smaller models became viable. That said, for something like this, I’d probably get more out of simply finding an existing implementation on github or the like and downloading that. When it comes to specialized and narrow domains like Space Invaders, the training set is likel…
Apart from using a Mac, what can you use for inference with reasonable performance? Is a Mac the only realistic option at the moment?
Most likely its training data included countless Space Invaders in various programming languages.
and probably some synthetic data are generated copy of the games already on the dataset? i have this feeling with LLM's generated react frontend, they all look the same
Earlier quoted context omitted.
True where trivial; where nontrivial, false. Trivially, humans don't emit something they don't know either. You don't spontaneously figure out Javascript from first principles, you put together your existing knowledge into new shapes. Nontrivially, LLMs can absolutely produce code for entirely new requirements. I've seen them do it many times. Will it be put together from smaller fragments? Yes, this is called "exper…
Humans can observe ants and invent any colony optimization. AIs can’t. Humans can explore what they don’t know. AIs can’t.
Nothing ultimately matters in this business except the first couple of time derivatives.
Earlier quoted context omitted.
The subjects in the study you are referencing also believed that they were more productive with it. What metrics do you have to convince yourself you aren't under the same illusionary bias they were?
Yesterday I used ffmpeg to extract the frame at the 13 second mark of a video out as a JPEG. If I didn't have an LLM to figure that out for me I wouldn't have done it at all.
Earlier quoted context omitted.
I'll believe they are doing that when one of the models draws me an SVG that actually looks like a pelican.
Someone needs to craft a beautifully bike donned by a pelican, throw in some seo, and see how long it takes a model to replicate it. Simon probably wouldn't be happy about killing his multi-year evaluation metric though...
My pelican on a bicycle benchmark is a long con. The goal is to finally get a good SVG of a pelican riding a bicycle, and if I can trick AI labs into investing significant effort in cheating on my benchmark then fine, that gets me my pelican!
> Two years ago when I first tried LLaMA I never dreamed that the same laptop I was using then would one day be able to run models with capabilities as strong as what I’m seeing from GLM 4.5 Air—and Mistral 3.2 Small, and Gemma 3, and Qwen 3, and a host of other high quality models that have emerged over the past six months. Yes, the open-models have surpassed my expectations in both quality and speed of release. For…
I'll bite. How do i train/make and/or use LoRA, or, separately, how do i fine-tune? I've been asking this for months, and no one has a decent answer. websearch on my end is seo/geo-spam, with no real instructions. I know how to make an SD LoRA, and use it. I've known how to do that for 2 years. So what's the big secret about LLM LoRA?
There are Colab Notebook tutorials around training models with it as well.
Earlier quoted context omitted.
> The real utility value of LLM for programming is to come up with something new. That's the goal for these projects anyways. I don't know that its true or feasible. I find the RAG models much more interesting myself, I see the technology as having far more value in search than generation. Rather than write some markov-chain reminiscent frankenstein function when I ask it how to solve a problem, I would like to see i…
"I would like to see it direct me to the original sources it would use to build those tokens" Sadly that's not feasible with transformer-based LLMs: those original sources are long gone by the time you actually get to use the model, scrambled a billion times into a trained set of weights. One thing that helped me understand this is understanding that every single token output by an LLM is the result of a calculation…
I tried to push for this type of model when an org I worked with over a decade ago was first exploring using the first generation of Tensorflow to drive customer service chatbots and was sadly ignored.
Earlier quoted context omitted.
Even "accidentally" it makes sense that "SVGs of pelicans riding bikes" are now included into datasets used for training as it has spread as a wildfire on the internet, making it less useful as a simple benchmark. This is why I keep all my benchmarks private and don't share anything about them publicly, as soon as you write about them anywhere publicly they'll stop being useful in some months.
> This is why I keep all my benchmarks private This is also why, if I were an artist or anyone commercially relying on creative output of any kind, I wouldn't be posting anything on the internet anymore, ever. The minute you make anything public, the engines will clone it to death and turn it into a commodity.
I considered experimenting with web DRM for art sites/portfolios, on the assumption that scrappers won't bother with the analog loophole (and dedicated art-style cloners would hopefully be disappointed by the quality), but gave up because of limited compatible devices for the strongest DRM levels, and HDCP being broken on those levels anyway. If the DRM technique caught on it would take attackers, at most, a few bucks and hours once to bypass it, and I don't think users would truly understand that upfront.
Earlier quoted context omitted.
Yesterday I used ffmpeg to extract the frame at the 13 second mark of a video out as a JPEG. If I didn't have an LLM to figure that out for me I wouldn't have done it at all.
You wouldn't have just typed "extract frame at timestamp as jpeg ffmpeg" into Google and used the StackExchange result that comes up first that gives you a command to do exactly that?