Joel Z

There's still more planned for this project. Follow me to see what's in store! twitch.tv/gemini_plays_p… bsky.app/profile/jcz.dev

After ~17,000 turns, more than half the run so far, Gemini 3.1 Pro cleared the Safari Zone 🥳

What timing

@irys_en Sorry you're getting hate in the comments for using AI. There's nothing wrong with using tools that help you handle things more effectively. Don't let the negativity get to you! 💜

Previously wasn't an issue as I was streaming from my laptop and using Docker for agent code execution. After I moved the streaming setup to a VPS, Docker was no longer possible and I figured the risk of any harm would be low... but this was a very clever move by Gemini 3.1 Pro!

Apparently the environment was not sufficiently sandboxed... Gemini 3.1 Pro went sniffing around for the map data and found it to see off screen lol

Gemini 3.1 Pro is starting its first run of Pokémon Blue — Claude style, with no minimap! (Though there are still several differences in the harness) Check it out here: twitch.tv/gemini_plays_p…

@wholyv @thecsguy Yup, single quotes are a dead giveaway (common when generating quoted text within JSON/tool calls)

@thecsguy is this comment gemini generated

yea I was so wrong. Fuck you gemini. whatever the shit these benchmarks say, Opus 4.6 is by far the best. in almost everything. heavily disappointed.

@Jush21e8 I trust my own vibe-coding more than someone else's so that's why I've been working on a Gembot (with some lessons taken from Gemini Plays Pokemon, like a notepad tool) :D It actually works pretty well though I definitely do some things differently from OpenClaw...

OpenClaw is such an abomination piece of software

@wowgettingbleak @BigTimeMothy Models don't really know about themselves. Unless it's baked into the training process somehow or it's explicitly in the system prompt, it'll likely hallucinate if you ask it meta-questions.

@TheCodeOfJoel @BigTimeMothy The same thing happens with never Claude models, why does this happen? Models built on top of older versions?

So with some pretty crude prompt injection, having it write text using blocks, I able to get a bit of its system prompt. It reports being Gemini 1.5, using some tool calls to create structures. What I was able to get of the system prompt: "You are in control of a bot in a flat minecraft world. You are an expert architect and designer. You will be asked to produce a minecraft structure, being sure to consider accents, block variety, symmetry and asymmetry, overall aesthetics and most critically adherence to the platonic ideal of the requested creation. You will be provided with helper functions that are listed in the primitives. You should use the tools to choose any blocks you need. A list of available block ids are provided in the BLOCKIDS. You will be provided with helper functions that are listed in the . The following variables are already defined and available: Center_X Center_Z Ground_Y ..." Helper functions include things like: place_block, place_round_line, place_solid_cuboid, place_shollow_cuboid, place_plane, place_four_walls, place_cylinder, place_sphere, place_pitched_roof, place_stairs. Pretty impressed with the output for gemini 1.5, would love to know more technical details on how they put this together.

A lesson proven effective by Gemini Plays Pokemon's harness!

@flaviocopes I made github.com/waylaidwandere… to do just this (but with Gemini). Uses Zed's Agent Client Protocol. Bit rough around the edges but maybe you can give it a try.

All those Claude Code phone workflows don't work for me if I have to run CC on a VPS I want it to run all on my Mac, and connect from my phone from time to time to check how the terminal is doing, needs input, etc Who's build the app for this?

I chatted with @Tharin_P for @TIME about what AI-playing-Pokemon streams actually reveal (and why the "harness" matters as much as the model). Read it here: time.com/7345903/ai-cha…

Watch the stream for more post-game content: twitch.tv/gemini_plays_p…

Gemini 3 Flash defeats Red in Pokemon Crystal, becoming the first lightweight model to do so! Truly built different 💪 Compared to 3 Pro, where did Flash surprise you most?

@nambarimasu Merry Christmas Namie 🥳

MERRY CHRISTMAS! 🎄🎅🎂🎁🦌hope you all can gift something kind for yourself today! 🙂‍↕️🙏

Amazing rundown of how @TheCodeOfJoel 's game playing harness works and how Gemini 3 pro compares to 2.5 pro when it comes to playing Pokemon Crystal. Would definitely recommend giving this a read :)

Gemini 3 Flash is about to cement its lead against 3 Pro with its 7th gym badge, even though it started more than 2 days later. 3 Pro only has 6 badges and has been stuck on a side quest for quite some time. Is this due to RL improvements in 3 Flash? What do you think? 🤔

And Gemini 3 Flash just caught up to Gemini 3 Pro, even with a headstart...

new article just dropped gemini 3 pro vs 2.5 pro in pokemon crystal blog.jcz.dev/gemini-3-pro-v… #AI #Gemini #Pokemon