Tom Armstrong

@NanoRaptor@bitbang.social Heh. This feels like some kind of mastodon serendipity:

@simon IMO, the problem with LLM code isn’t that its quality is *poor* but that it’s *indeterminate* in the absence of review by a skilled person. I’d say that for engineering purposes though, “poor” and “indeterminate” are close to the same thing. There’s that old principle that the cost of a bug increases the later it’s discovered. If“writing the code” is also the first review pass, then LLMs seem likely to push more bug discovery to further on in development.

@jesse@metasocial.com @evan@cosocial.ca @hongminhee@hollo.social @simon@fedi.simonwillison.net Many people I respect now use LLMs, people on this thread included. I use them too, but like the post author, I still get to decide when, and I try to choose carefully where to draw that line. If it’s important, or I’m learning something, I do it myself. The day my employer tells me I’m not using enough AI, I’m not sure I’ll know what my job is anymore. I used to pride myself on how good I was at training and mentoring junior devs, but nowadays, what juniors? 🫠

@Migueldeicaza@mastodon.social Excuse the potentially naive question, I’m not all that familiar with the context; the GitHub link implies that the breakthrough is training using the ANE, but the video shows inference, right? I had thought it was already possible to run the Qwen models on Apple hardware. What does the ANE reverse engineering effort do to aid the inference shown in th video?

@jesse@metasocial.com @evan@cosocial.ca @hongminhee@hollo.social @simon@fedi.simonwillison.net I feel so torn. LLM code tools are fun, and I probably enjoy them as much as anyone. But. I’m ~10 years into this job, and for most of them, everybody told me that the details matter, that understanding how it worked was important. Getting better at my job has often been synonymous with learning to pay better attention, learning how to get a handle on complexity, learning how to learn. These things generally feel at odds with use of LLMs, to me.

@luna@tech.lgbt I doubt it's peaked, either, even with all the existential threats we're currently facing. So, there are still Miles to go before we sleep.