It is an interesting essay. It leans on the side of "assistants are useful for simple coding tasks" and it's a bit more critical when it's about writing. The stance is original I find, yes it can help with some writing tasks, but if you look at the writing tasks you can expedite this way... if you wish to expedite them isn't it a sign that they were providing little value in the first place? Is the solution the assistant or changing the way you work? Indeed this might hide some busy work otherwise.

Interesting take on why people see more in LLM based systems than there really is. The parallels with psychics and mentalists tricks are well thought out.

This is how it should be done. This one comes with everything needed to reproduce the results. This is necessary to gain insights into how such models work internally.

Wondering how one can design a coding assistant? Here is an in depth explanation of the choices made by one of the solutions out there. There's quite some processing before and after actually running the inference with the LLM.

All the good reasons why productivity increases with code assistants are massively overestimated. To be used why not, but with a light touch.

AI supercharged scam. I guess we'll see more of those.

You should be mindful of the dependencies you add. Even more so when the name of the dependency has been proposed by a coding assistant.

Excellent work to improve Llama execution speed on CPU. It probably has all the tricks of the trade to accelerate this compute kernel.

Smaller models with smarter architectures and low-bit quantized models are two venues for more efficient use. I'm really curious how far they'll go. This article focuses on low-bit quantized models and the prospects are interesting.

Wondering where some of the biases of AI models generating images come from? This is an excellent deep dive into one of the most successful data sets used to train said models. And they've been curated by... statistical models not humans. This unsurprisingly amplifies biases all the way to the final models.

This is an excellent piece, I highly recommend reading it.

Interesting study on the impact generative AI can have on people performances in business settings. There are a few nuggets in there. In particular anything related to problem solving people do worse with generative AI tools than without. And even worse than that when they've been trained (probably due to overconfidence). The place where it seems to help is for more creativity related tasks... at the individual level, but at the collective level creativity decreases due to homogenization. Definitely things to keep in mind.

Very interesting piece. The chances that it is another bubble are high. It's currently surviving on a lot of wishful thinking and hypothetical. This really feels like borrowed time... I wonder what useful will remain once it all collapses. Coding assistants are very likely to survive. Clearly there could be interesting uses in a more sober approach.

Definitely this. It might ultimately impact the abstraction levels accessible to us for coding... but the skills will still be needed. Natural language is too ambiguous for the task.

Friendly reminder that the neural networks we use are very much artificial. They're also far from working like biological ones do.

This is one of the main problems with using those generative models as currently provided. It's time for the legislators to step up, we can't let a couple of players hoard energy and water for themselves.

Might be an interesting trick to reduce the computation and energy costs of large language models. Let's see if it gets replicated and generalized, this is a single short paper not peer reviewed anywhere as far as I can tell.

Interesting paper attempting to prove that hallucinations are unavoidable in those models. It is well balanced though, and explains why it's not necessarily a bad thing in theory. In my opinion, the problem is the marketing talk around those models making grand claims or denying the phenomenon.

Interesting tool for file type detection. Seems very accurate too.

Interesting paper evaluating a Life Cycle Assessment (LCA) method to estimate the power consumption and environmental impact of generative AI services. This is illustrated on a single service, hopefully we'll see more such assessments.

Very nice progress on this type of architecture. It's definitely needed in part because it lowers the inference cost quite a lot. It's also nice to see it released with under the Apache 2 license and the training set be documented.