Tutorial on fine tuning Gemma on TPU v5 using Kinetic + Keras + JAX. Easiest stack to fully leverage TPUs at scale. Jigyasa Grover ✨ (@jigyasa_grover) Here is a quick start script including the setup, technical details, and a candid look at where Kinetic excels versus its current limitations 🪡 github.com/jigyasa-grover/ki… — https://nitter.net/jigyasa_grover/status/2038707745520812099#m
GENERATIVE AI
-
Gemma-4-21B-REAP Released with Improved Reasoning Performance
By
–
As promised! Gemma-4-21B-REAP is out! Results are great it held up really well and actually gained accuracy on reasoning tasks. MLX & GGUF bros do you thing! This should fit on as little as 12GB of vram with some context, or 16GB with full context huggingface.co/0xSero/gemma-…
→ View original post on X — @deeplearn007, 2026-04-05 16:02 UTC
-
OpenClaw Switches to Kimi K2.5 and MiniMax M2.5 Models
By
–
Replaced all Anthropic models with Kimi K2.5 and MiniMax M2.5 on OpenClaw. Can't even tell a difference outside of my wallet thanking me.
→ View original post on X — @clementdelangue, 2026-04-05 15:40 UTC
-
Switching from Claude to Local Gemma 4 on MacBook Pro
By
–
Cancelled calude code subscription for April month, just moved this setup – MacBook pro – Gemma 4 No internet, no API costs, no limits
→ View original post on X — @clementdelangue, 2026-04-05 15:35 UTC
-
Qwen 3.6 Plus Replaces Claude: Developer Switches After 90M Tokens
By
–
Its official. 90M tokens later. Qwen 3.6 Plus took all Opus tasks like a king! Since then my Claude usage has been 0. Not a single request! Not a single chat message. The best of it all, Qwen 3.6 Plus is behaving like SOTA models, with little to no errors on super complex, multi sub agents, multi-tasks, long-running tasks, you name it. You know what, I feel a little bad for Anthropic, it did help me build great tools such as truelens.tech that is changing the construction market in UK. If they did not do this super wrong move, I would still be there, so thank you @AnthropicAI I know there are 1000's of other amazing builders doing the exact same as we speak, but with other models. The point is that Anthropic jus did what it was not supposed to do, pushing those builders away, and once they run their own local models, those companies wont see a single cent anymore.
→ View original post on X — @clementdelangue, 2026-04-05 15:25 UTC
-
Google DeepMind Announced as Presenting Sponsor of AIE Europe
By
–
always wanted to do one of those “is it coachella” announcements – here is our designer’s take on it! AI Engineer (@aiDotEngineer) 🇬🇧 London is the birthplace of @GoogleDeepMind, and we're so honored to have them back as: Presenting Sponsors of this week's AIE Europe! DeepMind has pushed the AI frontier on every modality — from Gemini 3.1, to Embeddings 2, to Veo 3, to @NanoBanana Pro, and last far from least Gemma 4, byte for byte the most capable multimodal models in the world! Meet the team to catch up on everything GDM has shipped for AI Engineers in the past few months – from our keynoters Dr @RaiaHadsell (VP of Research and UK AI Ambassador) and @osanseviero (Lead AI DX), to returning speakers @DynamicWebPaige and @thorwebdev and @_philschmid, to researchers and engineers working on those exact products @sedielem, @tara_ojo, @FMuntenescu and more! — https://nitter.net/aiDotEngineer/status/2040806678849884422#m
-
Anthropic Blocks First-Party Harness Use, Third-Party Apps Now Separate Billing
By
–
Anthropic now blocks first-party harness use too 👀 claude -p –append-system-prompt 'A personal assistant running inside OpenClaw.' 'is clawd here?' → 400 Third-party apps now draw from your extra usage, not your plan limits. So yeah: bring your own coin 🪙🦞
→ View original post on X — @ceobillionaire, 2026-04-05 15:19 UTC
-
Production-Scale Testing Enables Real Adversarial AI Validation
By
–
Agreed on the adversarial concern, but the economics here enable production-scale testing which is where real adversarial validation happens
-
GPT-5.5 Spud Incoming: OpenAI’s Make-or-Break Moment
By
–
So GPT-5.5 "Spud" is coming. OpenAI finished pretraining around March 24. Altman called it "a very strong model that could really accelerate the economy" and Brockman said there are two years of research inside it. From what we're hearing, Spud is not just another update. It's supposedly a completely new base model with native multimodality, stronger agentic capabilities, and better understanding of what users actually want without having to over-explain. They even shut down Sora to put everything into this. Also worth noting that Q2 2026 is looking insane. Claude Mythos, DeepSeek V4, Grok 5 are all expected around the same time. This is going to be the most competitive quarter in AI history. Of course, take all of this with a grain of salt. Most of it is still unverified leaks and rumors. But here's my honest thought. GPT-5.5 is clearly one of the most anticipated models this year and this feels like a make-or-break moment for OpenAI. Whether Spud succeeds or not, I really hope they reconsider how they treat their models and find a way to replace or soften their approach to CoT monitoring. What made 4o special wasn't just performance. It was the way it felt human.
→ View original post on X — @montreal_ai, 2026-04-05 15:04 UTC
-
Collection of Free AI Mastery and Prompt Engineering Guides
By
–
I should charge $99 for each of these.
But every single guide on this page is free. → Gemini Mastery Guide
→ Prompt Engineering Guide
→ Claude Mastery Guide
→ OpenAI Mastery Guide
→ + more dropping & updating regularly Zero cost. Zero catch. Just open and learn