Cool project, comes with a little too much guidance in the harness though IMO (pathfinding, textual milestones etc). (The author is very upfront about this in their README though)
I think if it was combined with a regular vLLM it could be really interesting, especially watching the reasoning logs.
Bonus points if it was one of the latest open models that somehow had all prior training knowledge of Pokemon abliterated so it was reasoning as an intelligent persona that had no knowledge of even the concept of Pokemon.
I also tried to create a “Jev plays Pokémon” but with minimal additional context outside a move history and what is available in memory from the emulator, so no pathfinder, predetermined game path, etc. I can safely say that this experiment failed however, and Jev was not able to even get to Professor Oak’s lab to get a starter.
Props to OP for getting a working version, but it does not seem that this model is capable enough to play Pokémon at this point in time.
Thank you. What I realized is that Jev is incredibly powerful for the decision making part of it when presented with a scenario that gives it a little context. It looks like we’re gonna be able to beat this game for under $2 of tokens. This is what blows my mind
I guess the difference is the folks on the 100 year ship won’t get any feedback that the tech leapfrogs have happened, whereas at least some of the software players have a chance to use their market knowledge to at least adapt.
The marketing outreach in general for too many companies still leans heavily on “we use AI”, whereas most customers just want results and whether you use AI to either provide those results (or to make it cheaper or faster) isn’t a primary concern of the customer.
Some countries have more complexity to registering a new company than others, and perhaps it’s the establishment of a new legal entity that is an unlocker to the free credits.
I don’t know if that’s true of India or not, but I like to give comment authors the benefit of the doubt that being specific about the geography was helpful context here
India is very bureaucratic and slow for starting a new business new entity. It ranked 63rd on the most recent world bank survey of 2020[1], and it is even more painful to close one.
So that report was discontinued after relatively proven statistical gaming and interference by several countries, doesn’t exist post 2020.
Post 2020 SPICE+ launched making registration trivially fast. However really the trick isn’t creating and closing companies, it’s that the fact majority of reporting that would catch scam companies like these just get ignored and gather small civil penalties payable years later or not at all, while keeping up with major reporting that might close their bank accounts… it’s a very disorganised system that facilitates a lot of unpleasant activity.
I didn’t make any claims towards spesific scams, just a lack of effective mechanisms to hold companies accountable, a lot of countries governance systems do indeed have audits and mechanisms to address companies making large numbers of individuals unhappy or where large number of individuals feel they are being treated unfairly.
Honestly if company size of Amazon is not doing their due diligence that is their own fault. It is not that big of ask for them to spend say a day or week evaluating these things.
The world bank is best quality source I could find, if you know a better source or any fact contradicting the claim India is high friction for general ease of business, that would be welcome.
Quick question, did you use an LLM to help with this? (Not judging one way or another).
The only reason I ask is that for the past couple of weeks I’ve been making a similar utility to test out the Sol model and it’s hilarious how similar the architecture and even the CLI switches are to what Codex came up with.
It didn’t come off as dismissive to me. I was curious as well as to where such optimizing helps and knew that the answers to your question would help me discover use cases I didn’t think of
Moving a lot of files from old laptop to new, lots of duplicate videos in scattered directories with different file names.
Point the CLI tool at a directory and up pops a ncurses application that scans from the directory you specify (recursively), hashes all the files to get groups of dupes, and then ranks the dupes into a table (that looks like top) so you can work your way through the dupes and decide which to keep without your hands leaving the keyboard.
It automatically ignores typical developer directories like node_modules so you're not tortured with noise in the result set.
Hit 'p' and a preview pops up so you can double check they are indeed dupes, if it's a media file the preview window autoplays the dupe videos in tiles with zero volume.
Supports a dry run mode, and switches to cover behaviour like sending dupes to trash rather than hard deletion.
Yeah if I need to I take both also. In addition I be sure to have a caffeinated drink also as caffeine has been shown to both speed the absorption and boost the efficacy (5-10%) of paracetemol over a multi hour period. https://pubmed.ncbi.nlm.nih.gov/17442681/
Exactly - I'm thinking the bad spatial navigators have a higher probability of washing out of driving and pursue some other career. They may not say "I'm bad at figuring out where I am", but the economics of the job are just a little bit worse for these people.
I think if it was combined with a regular vLLM it could be really interesting, especially watching the reasoning logs.
Bonus points if it was one of the latest open models that somehow had all prior training knowledge of Pokemon abliterated so it was reasoning as an intelligent persona that had no knowledge of even the concept of Pokemon.
reply