Harrison Kinsley

Trying to get something from epic tech demo that runs on individual machines to an actual first release is heckin humbling This is just the beginning of what we're cooking, I cannot wait to share more Use it, tell us what you like, hate, & want. My DMs are open & we're hiring!

@__K8r @ApplyWiseAi I do not think closed source has actually been progressing very well at all tbh. Harness progress has been huge. Actual model perf is p negligible imho. Are you referring to some specific benchmark? or are you materially getting staggeringly better perf from your work? I'm not.

@Sentdex @ApplyWiseAi closed source has been progressing pretty well. you're saying they hit a wall? I think you said that a while ago

While closed source AI is in shambles, open source is having one of the best weeks of all time. Z ai GLM 5.2 Minimax M3 Kimi 2.7 code

@JeffSte17327059 i have full faith in GLM that they will post the weights when they said they would

@Sentdex only kimi and minimax have posted their weights on huggingface not glm

@ApplyWiseAi ok actually 3rd option is both. Go public with a tiny sliver of total shares, but at massive market cap valuation for big raise. Then nationalization via government stake at inflated prices. Gov't backstops price and inflated valuation this way.

@ApplyWiseAi Closed source is most definitely in shambles. OAI and Anthropic have 2 options for survival rn: Go public at bubble prices and raise next rounds by requiring the bubble keeps growing significantly, or die. Become nationalized and too big to fail. Those are the options.

@SearchDataEng @ApplyWiseAi I think actually itll be a combo of things. Go public, then gov't takes a stake, providing a liquidity backstop.

@Sentdex @ApplyWiseAi What might Nationalization look like, is there precedence for such an event?

@InnoboSJ @ApplyWiseAi although maybe the real play is is both. Go public then nationalize with a govt stake. Then the govt just simply backstops as liquidity to prop up the company. nvm i decided this is exactly the path they'll take.

@InnoboSJ @ApplyWiseAi I think OAI and Anthropic leadership both probably prefer nationalization. Going public just kicks the can down the road.

@leepy_cat Essentially "deep research" and you really dont need much intelligence at all to do that. I just prefer if we can knock the middle man out of what we're searching for.

@Sentdex anecdotally, i've been using the new siri and its great at searching web and summarizing which is the bulk of the casual usecase 😭 its amazing that we're at the mobile + free + pretty good part already

@Arm4x x.com/Sentdex/status…

@Sentdex you forgot prefeitura-rio/Rio-3.5-Open-397B

@joaomviso im a wee bit sus of anyone who comes out of what feels like no where with a massive banger release. I don't think I've ever seen it be real. I want to believe. But I'm holding my hype for this model atm.

@Sentdex and Rio de Janeiro city qwen 3.5 achieving gpt 5.5 levels: huggingface.co/prefeitura-rio…

@JessieTweeting @cohere lol banger tweet from them, but I think their models are still a wee bit behind. but they're on my radar 100%. If I'm hype about a model, it means it can likely replace claude or oai usage. Not seeing that for cohere's models just yet.

@Sentdex Don’t leave out @cohere 🇨🇦 x.com/cohere/status/…

You can continue to use Command A+ and North Mini Code whether we want you to or not

@hansleyc 3-6 months behind max IMHO.

@Sentdex They are months maybe one year back

@vildanden_ai yes. M3 is a good model. No idea where the M3 hate is coming from. I'm using it and it's an epic release.

@Sentdex bro M3 😅😅😅 seriously

@DustdrifterX M3 is better than 2.7. No question. IDK why people say M3 sucks. It's objectively better on any test and this comes through in my usage. Also comical, because when it was m2.7 people shat on that model. Now it's the good ol times lmao.

@Sentdex tbh Minimax M3 kinda sucks, 2.7 is really good but too slow

@Stumblinz Using M3 locally, the only one of these that I can actually run. So far, it's very nice. llama.cpp doesnt fully support it yet so it's still kinda goofy with toolcalls. I will definitely be giving Kimi 2.7 code a try thru the @togethercompute api, and 5.2 once it's on OR API.

Have you tried any of these yet? How are they? I tried Kimi 2.6 a weekend ago and I let it build an MMORPG zone and it wasn’t the worst, but for the money I felt safer with Anthropic. I think though that Kimi felt maybe a few iterations short of taking my full attention. I’d give it a few months before maybe I’m using them full time but in a private cloud

@yuvalav couldnt figure out how to properly patch llama.cpp and hermes to get m3 to be reliable unfort. but it was easy and p fun to get things going with my own agent

@Sentdex Finally!!! this sounds really cool! but I didn't understand why did you end up spending time to apply fixes if the minion can improve itself and solve issues? couldn't the minion apply the fixes?

After spending too many hours trying to implement fixes for MiniMax M3 native tool calls serving via llama.cpp to work in existing agents, I simply had M3 write its own mini coding agent I'm calling: Minion Now my minion just edits itself to give me what I want as a coding agent and it works surprisingly well. Lots of changes I plan to make, feel free to use it if you like...but mostly it has me questioning if we all should just make our own agents at this point. Maybe MiniMax is exceptionally good at tool calls out the gate to make this super simple, but I am enjoying making my minion exactly what I want and nothing more! and it doesn't take 50K context to say "hi" (yet) We'll see how long I can keep my own bloat at bay. Also MiniMax M3 overall so far has me very impressed. This is a VERY cool model!

@M_Chimiste @UnslothAI @MiniMax_AI I dont know if it'll work ootb wit anything but minimax m3 but shouldnt be too hard to allow for others. M3 has some special tokens stuff goin on that im deliberately handling for. Eventually ill make it function with others too maybe. Rly just wanted to play with m3 lol

@Sentdex @UnslothAI @MiniMax_AI I’m going to fork and also add code2lora. Will report back.