I mostly use Opus for skill development. Once I have a solid skill implementation with a good eval, I move ongoing execution to a cheaper model running under Goose. With the eval you can see if the cheaper model works well enough.
It was odd that there was no mention of the forced adaptive reasoning in the article. My guess is they don't have enough compute to do anything else here.
I’ve only played with Junie and Aider so far and like the approach the Junie agent takes of reading the code to understand it vs the Aider approach of using the repo map to understand the codebase.
Thanks for sharing. Can you add a simple example that is more concrete and less abstrwct to the docs? It would be easier for people to understand if you did.
I’ve found the OpenAI assistants API not really up to snuff in terms of predictable behavior yet.
That said, I’m very bullish on agents overall though and expect that once they get their assistants behaving a bit more predictably we will see some cool stuff.
It’s really quite magical to see one of these think through how to solve a problem, use custom tools that you implement to help solve it, and come to a solution.
This right here is actually the coolest part about developing with LLMs. You just changed the functionality with a sentence rather than a config file, or writing code. It’s great to be able to break out functionalty into things that can be easily handled in English (or your human language of choice) or what should be done in code.
That’s right. It was selective availability and you used differential post correction to clean the data up and get accurate locations for the data you were capturing. Thanks for the correction!
The interesting thing about GPS back then was that the location data you got back were randomly slightly incorrect every time you got a reading. I’m pretty sure this was so it wasn’t useful for military purposes.
I think it was called differential post correction but if you had a base station with a known location you could snap your incorrect points to the difference generated at that correction level and get the true location after the fact.
Source: GIS major in late 90s when this stuff was a lot more magical
Don’t forget classification. Even the zero shot classification can be quite decent, but put a solid prompt together, or fine tune a set of examples and it’s way easier with an LLM than previous methods in my experience.
Maybe I’ve been using PostgreSQL too long but when faced with the choice of adding vector support to PostgreSQL or using a new technology, my first choice was to start with the PostgreSQL addition.
I’m not criticizing the specialized case for a true vector database, but for most workloads I agree that the big database players will be the right choice for many users.
I’m fine if you have a “contact us” for a true enterprise plan, but companies that have actual pricing plans should post them. I way prefer to buy from SaaS companies that display their pricing up front rather than trying to get me on the phone to talk with an account executive.
I’m not scared to pay serious money for a service, but putting your service behind a sales person is more than likely going to cost you my business.
What I find entertaining/confounding is how difficult the abstracts to these new AI papers are to understand. It feels like academia is pushing this style, so it’s hard to blame the authors since they have to play the game.
For reference I have an undergrad degree in computer science, have been working professionally for 25 years, and am fairly data centric in my work.
I’m hoping when I run this through GPT4 to get an explanation for a mortal software developer something sensible comes out the other end.