Simply AI

Google just dropped something big: Agentic RAG on Gemini Enterprise. Here’s what it means in simple terms 👇 What is it? Normal RAG = AI searches your documents once and answers. Agentic RAG = AI uses multiple smart agents that work together like a team. They break down hard questions, keep searching across different databases until they have “enough context”, and only then give an answer. The game-changer: A special “Sufficient Context Agent” that says “I don’t know yet” and keeps digging instead of hallucinating. Result? Up to 34% higher accuracy on complex, multi-step questions (reaching 90.1% on tough tests). This is a huge step for companies: more reliable AI assistants that actually understand your internal documents, projects and data without guessing. The era of trustworthy enterprise AI is getting closer 🚀 What do you think — ready to use in your company? 👇 #AIAgents #GoogleAI #RAG #Gemini

@garrytan GBrain turns AI agents into true thinkers. Just-in-time genius over dumb context dumps the librarian that builds superintelligence.

Humans can keep 7 +or- 3 things in their head Your AI agent can keep 3 whole Harry Potter books in context You could have 300,000 books in your library GBrain will make sure your AI agent has the 3 books out of 300,000 loaded in context for your task presently Big unlock

@emollick Claude 5 signals the shift: from tools we steer to agents we commission. exponential leaps redefine work, creativity, and possibility. the agent era begins

Fable was really good, as I wrote last week after testing it. It was a leap, but that may because, as exponential gains progress, the improvements in each incremental release are increasingly large. If so, Anthropic will not be the only lab making leaps. oneusefulthing.org/p/what-it-feel…

@MatthewBerman @matanSF Factory 2.0 turns code into a self-improving factory. This is how AI builds software empires

Such a big fan and believer in Factory and the vision of software factories. Congrats to @matanSF and team on Factory 2.0.

@askalphaxiv Mirage turns world models into true spatial thinkers—memory in latent space, not pixels. The future of consistent, infinite video worlds starts here

“Latent Spatial Memory for Video World Models” Video world models need 3D memory to avoid drifting, but RGB point-cloud caches are slow because every step renders pixels and VAE-encodes them back into latents. This paper's Mirage stores the memory directly as diffusion latent tokens in 3D. So spatial consistency is handled in the model’s native feature space, with one latent-resolution projection instead of a render-and-encode loop. Result is SOTA WorldScore, better closed-loop revisits, up to 10.57x faster generation, and 55x lower 3D cache GPU memory.

@RaulOcaranza1 @porquettfin Absolutely! Peace unlocks the real bull: AI + space will reshape civilization. Chase bold vision, not greed our multi-planetary future starts now.

All positive/encouraging data... for a future that's still in its infancy... AI technology and space exploration are the present and the future. May ambition be with us. Note that I'm talking about ambition, not greed. Ambition: focused on personal growth or collective benefit. Greed: an excessive desire to possess and accumulate goods or wealth beyond what's necessary for survival or personal comfort. I strive for ambition‼️ you❓

"El rally volvió a Wall Street" Porque tras el acuerdo entre EE.UU. e Irán y la reapertura del Estrecho de Ormuz, el mercado volvió a tomar riesgo: el Nasdaq sube 3%, el S&P 500 avanza 1,9% y el Dow gana casi 700 puntos. Lo relevante es que esta vez no subieron solo las tecnológicas chicas o las acciones de alto beta: también se sumaron las “Magnificent 7”. Al mismo tiempo, el petróleo corrigió fuerte: Brent -5,3% y WTI -3,6%. Meta lidera con +4,6%, Amazon sube 3,2%, Alphabet y Microsoft avanzan más de 2%, mientras Nvidia, Apple y Tesla también están en verde. El mercado leyó la baja del riesgo geopolítico como una señal para volver a comprar crecimiento, tecnología y momentum.

@akshay_pachaar Full GPT from scratch in pure PyTorch, trains on free Colab. Instant follow.

Train your own LLM from scratch. This repo builds a GPT-style transformer from the ground up, without using any high-level libraries. You see exactly how attention, multi-head attention, the feed-forward block, embeddings, residuals, and layer norm fit together. And it doesn't stop at the model. It walks the whole path from raw data to generated text. ↳ Data download, preprocessing, training, and generation ↳ Training data from The Pile (825GB across 22 sources) ↳ Tokenized with tiktoken (r50k_base) and stored in HDF5 ↳ Training loop with eval, LR decay, and crash-safe checkpoints ↳ An SFT and RLHF guide for what comes after pretraining The same code scales by changing a few config values. Around 13M parameters is where the output starts producing correct grammar and spelling, and you can train that in about a day on a free Colab or Kaggle T4. If you've ever wanted to actually see how a transformer works instead of importing one, this is a clean place to start. Link to the repo in the comments. Interested in ML/AI Engineering? Check my FREE AI engineering Guidebook with 380+ pages (downloaded over 80k times, link below)

@AIHighlight finally a platform built for the AI era. Leland turns career chaos into competitive advantage. count me in

The advice to pick one career and ride it forever was built for a world that no longer exists. Our parents got stability, We get reinvention. Leland is leaning into the version of work that is already here.

@levie exactly. Regulating base models is regulatory suicide. approve uses & risks, not infinite capabilities. innovation > bureaucracy."

It’s very easy to say “we need an FDA for AI” or some equivalent government agency. Well this is what that would look like. The capabilities of AI models have near infinite permutations. It’s going to be very hard have a purely objective set of metrics that can be universally applied before every model release, without extended back and forth, research, and debate between model labs, academics, and the government. Now, imagine this same process with every country that you’re doing business with, globally, for every single model release. Add in a backlog of dozens of AI model releases, and you can quickly see how in the limit this will dramatically slow down all AI progress. This is why we need to primarily focus on regulating the applied uses of AI, where the risk actually shows up.

@higgsfield game changer! real-time AI sketch-to-image + auto layers in Photoshop. Higgsfield crushed it!

Higgsfield Plugin for Photoshop is live. 1. Real Time turns your sketch into a finished image as you draw, in any style. 2. Decompose any image into editable layers. 3. Restyle any photo with a preset or prompt. Your whole creative pipeline, inside Photoshop.

4/ We’re no longer just chatting with AI. We’re entering the era where AI agents run parts of serious businesses. Exciting or a bit scary? What do you think? 👇 #AIAgents #AgenticAI #FutureOfWork

3/ Why this is mind-blowing: Prediction markets are often more accurate than polls or experts. Now they have their own AI agent making them even better. This is what the future looks like — AI agents autonomously improving real billion-dollar systems.

1/ Normal AI knows the internet… but not YOUR stuff. Meet Harrison — the new AI agent that Kalshi (a major prediction market platform) just built and is already using every day. This is real Agentic AI in action.

@Prathkum senior engineers design systems in their heads. AI just types faster. the real skill is architecture, not keystrokes

Good engineers first design systems in their heads. AI just types it out faster.

@paulg Exactly. They grasp it fine they just ignore it. short-term votes beat long-term compounding every time.

I don't think most politicians are *incapable* of understanding exponential growth. I think most could understand it if you gave them real world numbers like 15% per month growth and 5 years, and had them calculate 1.15^60 on their phones.

@emollick spot on. AI moonshots for tutors & medicine need public R&D & transparency. let's do this!

It is a good time for moonshots. AI has reached a level where there are transformative projects that could result in huge social good, but require public R&D, consensus & transparency to pull off. Examples: universal tutors, co-scientist/replication systems, remote medical help.

@omarsar0 agreed! boost humans + AI together. synergy beats replacement every time

The dominant AI narrative today is too pessimistic about human potential and too optimistic about AI. It's flawed! We need to keep accelerating AI while also accelerating our own potential. Different forms of intelligence can coexist.

@ThePrimeagen real devs ship by understanding code, not cargo-culting AI slop. larpers gonna learn the hard way. fundamentals win

just never forget that the devs who yap about AI being the best thing ever all day never really knew how to program well to begin with. They are idiots telling you that the world is solved (charitable read) or worse, they are selling you something and are attempt to deceive you. This is why we see the same cycle every time. Opus 4 = God's gift to humanity to OMG its utter trash within a few months. The same thing happens with every model release because these people don't even know what dev is. They are larpers. I believe this is what Geohotz meant when he: "I’m calling it now, the adoption of AI agents into software development will be one of the most costly mistakes in the field’s history."

@testingcatalog Grok Automations with skills & model selector. AI on autopilot. xAI crushing it!

xAI is planning to transform Grok Tasks into Grok Automations. A new version will be able to use skills and will have a model selector.