20+ years ago I always did our final interview, which was 4-6 hours on one of a handful of technical topics, picked to be something that I could see the candidate wasn't previously familiar with. It was a good test of tenacity, curiosity, and open-mindedness, not just pure tech
@jeremyphoward
-
Learning NLP Foundations and Recent Approaches with Karpathy
By
–
They're the best way to learn the foundations IMO. I'd also look at @karpathy
's recent videos to learn recent NLP approaches. -
Claude-Specific Context Guide and Best Practices Reference
By
–
It is! Give it this context — it's particularly designed for Claude: https://
fastht.ml/docs/ref/best_
practice.md
… -
MoE Performance: Discrete GPUs vs Unified Memory Architecture
By
–
MoE is strictly worse for folks using discrete consumer GPUs like a 3090 AFAICT. It seems pretty good for unified memory stuff like Macs though.
-
Learning AI Foundations While Leveraging Practical Tools
By
–
Not a dumb question at all 🙂 I think learning the foundations is still just as important, and fastai is still the best way I know. But you should *use* AI a lot too.
-
Semi-Supervised Learning Tasks for Computer Vision and NLP
By
–
If you're looking for inspiration, I have quite a long list of semi supervised learning tasks for computer vision, many of which have natural analogues in NLP. Here you go —> https://
fast.ai/posts/2020-01-
13-self_supervised.html
… -
Communicating AI Capabilities Versus Limitations
By
–
From knowing both you guys a bit, I'm not sure you actually have very different opinions about the capabilities or what they imply — but just different opinions about whether the most valuable communication at the moment is to explain the limitations, or the capabilities.
-
LLMs Writing Better FastHTML Code Through Documentation Improvements
By
–
Yes lots. Spent a long time testing how well models could write idiomatic FastHTML whilst tweaking each part of the docs. Generally when LLMs did poorly, it turned out our docs for humans needed improvement too! 🙂
-
Quality Digital Stimulus Better Than Screen Avoidance for Kids
By
–
No I think you're misunderstanding my point, which is people (kids and adults) do better when they have more opportunities, skills, high quality stimulus, etc. Yes, passive asocial non-educational entertainment is generally bad — so let's say *that*. Not "no screens; be bored"
-
GPUs throttle to low idle power when inactive
By
–
GPUs throttle down to low idle power when not actively doing calculations BTW.