This was about a year ago, but I had trouble getting podman to add an nvidia GPU so that the container could use it. It was technically possible (I succeeded) but it was annoying and "different".
I'm curious what you think of as "the mean"? I consider the input training set for an LLM to contain its mean. My hypothesis would be: an LLM alone cannot consistently produce code above the mean of the quality it was trained on.
> LLMs are regression-to-the-mean machines--they pull junior developers up, and drag senior developers down. Taming them requires trading the romance of 'code as craft' for the physics of manufacturing.
The thing I don't know is: how do we decide which direction is most valuable? I can see arguments in both directions--quality vs quantity, essentially. I think there's a strong argument for the value of both:
- we need more quantity of software: for a long time, the ability to write software has been locked up, confined to a closed cabal of specialists
- we need more quality in software: we depend more and more on software in every aspect of our lives, mistakes are intolerable and should be avoided
I think it depends on which side of the regression-to-the-mean machine that you land on (above or below the mean) for any given skill that is being disrupted by AI. From above, AI is frustrating; from below, it's magical.
The frequency of choosing to go out to the movies is also about how often I think "I wish I could do this in VR".
Examples:
- Before going on a trip, pre-visiting the destination in Google Earth with VR is very spatially informative & makes directional intuition memorable upon arrival at the real world destination.
- Virtual role-play with environmental cues that cause make-believe to be ever more real.
But most people don't need this very often. Picking up a book or throwing on some earbuds to listen to a book are far more frequent and compatible with simultaneous other activities. VR feels the same--a high-demand focused experience that is infrequently worth the effort.
I like the concept, but I bailed at "GPT 5". The only thing that has given me peace of mind and the ability to journal honestly and successfully is Obsidian, because it lets me own my data (as text files).
FWIW as someone with only a pinky toe in the Zig community, it's quite engaging and interesting to see a blog post like this. It makes me want to learn more, and reminds me that there's a wide tent here (that might even include me!), not just a tight-knit "inside" group.
I don't know the bank they are referring to, but I can cite an example for me: RBC Royal Bank of Canada requires the mobile app. There is nothing you can do on their website without first 2FA via their specific mobile app, and even then only in limited transaction sizes. If you want "full access" (e.g. up to $10k daily transfer via e-transfer) then you MUST use biometrics and the mobile app.
Seth Kaplan, professor at Johns Hopkins University and frequent contributor to UN and World Bank efforts to shore up community in difficult countries has written a book called Fragile Neighborhoods that I highly recommend.
I built a spatial platform similar to gather.town called relm.us in 2021 (now MIT-licensed open source [1]) and was hired by an edtech company in 2023 because of the expertise I'd gained in overlaying audio/video for participants in the game world.