Lachlan

One of my favourite ways to get a vibe for AI models coding ability (particularly it's ui preference) is via @lmarena_ai they have their WebDev arena which pits two AI's against each other, blindly and you pick which result you like the most. You begin to learn the styles that particular AI uses, it helps you understand why people generally prefer Anthropic outputs. Both their 3.5 models, Sonnet & Haiku share a similar design style which is more aesthetically pleasing for sure. I was also impressed with some of the new Gemini 2.0 Pro outputs, and Qwen Max, these are definitely a leap up compared to previous gen models also. But still just behind sonnet 3.5 overall, generally* I was generally underwhelmed by the O3 mini high outputs, (particularly UI). *It's a very useful exercise to show you that there isn't one "best" model, even though it's easy to say "Claude is the best" - Each prompt is different and getting opinions from other models is very useful.

@inductionheads Nah look outside coding and the gemini models have been pretty decent. Solid in voice, image, video TPUs power most things at 1m context Money to burn competitors inference compensation budgets into the ground. own 16% anthropic theyll be fine / do well

And you’re bearish on Google?

@vineettiruvadi @0xSero You can, you just need a shitload of vram which most local setups don't have.

@0xSero am I missing something or is GLM 5.2 not runnable on local hardware?

AGI acheived internally.

Fable with medium or high thinking and /goal, avoid ultracode / workflows if you want to not blow through your usage in one request.

@theo #freefable

Did you know it’s possible to summon hundreds of people with crippling AI psychosis on this site? All you have to do is add a hashtag to your post. #keep4o

Interesting time for Google & OpenAI, if you did have a Mythos level model ready to ship, would you ship it? Or wait until Anthropic has gone through the mess with the government for more clarity.

@ThePrimeagen Yeah not sure when it turned but has been absolutely horrible for a while now.. I had to turn auto-correct off, their keyboard is still whack somehow.

How does anyone use an iphone? The keyboard and how you interact with text is awful. So so very awful

@midjourney Medjourney

Announcing a new division of Midjourney called "Midjourney Medical"

Medjourney

@midjourney Medjourney

A technical dive inside our new "Midjourney Scanner"

@briandecoded @petergyang /goal create a new operating system for linux, make no mistakes - should be done in 30 secs right.

@petergyang It shouldn't be running for 2 hours... Sounds like you didn't sandbox it properly. I have it running on /goal and it takes about 30 seconds.

So I have Codex running on a /goal and it's been working for 2 hours but the problem is it's making alot of wrong assumptions so I have to monitor and steer it constantly. Is this expected? Perhaps I should've had it make a detailed plan first?

@joshcirre Your settings? 😂

Oh no. What have I done? I will never financially recover from this.

@_imdawon No idea who is in charge of their search but they gotta go. So annoying haha i don't want to have to install raycast to find a fcking application.

macOS Tahoe was such a mistake i still cant search for applications theres no OS for the working man anymore

I quite like the model its performing quite well in our harness compared with the preview model, however the 3x in price is hard to justify / changes the utility of the model significantly. if the pricing hadn't been so much higher i think the reception from people would have been a lot better.

Gemini 3.5 feels like the start of a new era for Gemini, we spent the last 2.5 years putting the infrastructure, products, team, etc in place (learning lots of lessons along the way). The model is the product, please keep the feedback coming!

@zachtratar Yeah pretty crazy how many security issues we are seeing. Was just thinking the same thing - x.com/lowcodelocky/s…

With the amount of supply chain attacks across different packages developing with Xanoscript keeps making more and more sense.

@swyx /goal keep working until I say stop..

@samhogan Nice work Sam, having a play now!

We’re introducing HALO 😇 Hierarchal Agent Loop Optimizer HALO is an RLM-based agent optimization technique capable of recursively self-improving agents by analyzing their execution traces and suggesting changes. This work is inspired by the Mismanaged Genius Hypothesis proposed by @a1zhang and @lateinteraction earlier this month. tldr; we improved performance on AppWorld (Sonnet 4.6) from 73.7 --> 89.5 (+15.8) by giving HALO-RLM access to harness trace data and asking it to identify issues. The feedback from HALO surfaced failures in the harness such as hallucinated tool calls, redundant arguments in tools, refusal loops, and semantic correctness issues. Each issue mapped cleanly to a direct prompt update. We then fed these finding into Cursor (Opus 4.6), and asked the coding agent to update the underlying harness. We repeated this trace -> HALO-RLM analysis -> code update loop until the score plateaued. Today we’re open-sourcing the core HALO-RLM framework, evals, and data for further review.

@matvelloso Agree it's a painful process - surprisingly easy for google though via their @GoogleAIStudio they integrated their AI auth / db services quite well.

I invite you to try to create a simple, functioning OAUTH app on both Google's and Microsoft's identity systems. Just that. Start with nothing, and count how long it takes.

@ivanfioravanti @0xSero yah sucks at coding unfortunately

@0xSero But in coding... better to avoid it as much as possible.

This is god tier and fits on 24gb + It crushes everything up to 6x its size and ties with Gemini deep think and Deepseek on Math huggingface.co/nvidia/Nemotro…

Going to be a fun week of launches : )