nbuujocjut

GitHub Copilot and similar tools make good developers more productive. This alone is a genuine use case with some associated value. Is it enough to justify the valuations of OpenAI etc? Probably not by itself. But I expect other industries have similar productivity boosts where people learn to use the tools appropriately.

What’s the total business opportunity of making all knowledge workers 10% more productive (to pick a more modest goal than outright replacement)?

Related: https://www.mjlivesey.co.uk/2025/02/01/llm-prog-lang.html

It’s doesn’t matter if a minority of passion techies will still be up for new tech, if the average developer just wanting to get the job done and relying on LLMs finds it harder, it will be a significant barrier.

Asking Claude this morning. Seems pretty reasonable and contains the warning about accuracy.

> Michael P. Taylor reassigned Brachiosaurus brancai to the new genus Giraffatitan in 2009. The species became Giraffatitan brancai based on significant anatomical differences from the type species Brachiosaurus altithorax.

> Given that this is quite specific paleontological taxonomy information, I should note that while I aim to be accurate, I may hallucinate details for such specialized questions. You may want to verify this information independently.

Ok so more about what is persuasive than what can trick the LLM?

Interesting challenge! Is the motivation to discover effective prompt injection techniques?

I agree that a lot of human output is stochastic parrot-like. But now and again at least, some human or group of humans appear to come up with something truly original and brilliant. This is something I have yet to see from LLM-based AI, and my guess is it would take a different paradigm to create machines that think in this sense.

I prefer to think of it as “test before commit” or “test before push”.

If I create a pull request with well written code and tests to match, and you can’t tell which came first, the code or the tests, it doesn’t matter in what order they were created.

At least here in Denmark it is very common to start with a bike that is build without pedals (and chain etc). It works really well, one of my kids could ride her pedal bike before 4 years old thanks to starting on a “løbecykel”.

How would you handle the situation that if this is successful, you’ll get massively more than 10 candidates each day?

Will most organisations be building out their own AI Infrastructure? I would guess in most cases they would either use APIs to existing models (OpenAI etc) or use cloud providers where they need to run and train their own models. Is there a risk that cloud providers cannot meet the demand?