Supernet AI 🌐

SuperNet is excited to announce that our core context memory technology Atomic Memory is live on GitHub today ⭐

@haider1 This also explains why users can have wildly different experiences with the same model depending on workflow and settings.

OpenAI Noam Brown says single-number benchmarks no longer make sense for modern AI models Once models can use CoT and extra inference compute, their performance depends heavily on how much reasoning time they are given "it made sense for GPT-2/3/4, but not for reasoning models"

@nivi the “right button” exists because researchers spent years building models capable of navigating those solution spaces in the first place.

When AI solves a new mathematical problem, it’s not magic. The solution was already in the algorithmic scope of the model—someone just had to press the right button.

@0interestrates because writing feels tied to human intent, while code is mostly judged by whether it works or not.

why do people (including me) have an aversion to AI writing but not as much to AI code? if a piece of text smells AI i stop reading it but i use things coded entirely with AI every day

@eigenrobot Honestly not that crazy of a take because AI will eventually become a layer embedded into almost every form of work & human systems in general.

long term i'm guessing AI ends up easily bigger than the internet, probably bigger than writing, somewhat lower than language

@LottoLabs Makes sense cause the value is shifting now from single responses to long-running autonomous workflows.

Qwen is very smart to go all in on agentic models

@boneGPT Honestly Gmail should be one of the best showcases for persistent AI memory and personalized workflows by now 👀

You need to stop benchmaxxing and start productmaxxing. Why compete in their arena of gameable benchmarks at all? Nobody cares how it did a percent better on AgentBenchBotMaxArcAgi3. Play in the arena you win. Example: Why isn't AI Gmail solved yet??? I sometimes get an autofilled post. Sometimes I don't. None of it is clear. Why can't I add memories and rules to my email? Why aren't there AI templates I can choose from for every reply that are based on what I would send? Who the hell is in charge over there? This should be the FIRST THING. I should have 3 response templates autogenerated for every email to pick from. I should have rules and memories. It should be obvious. Focus.

@Dr_Singularity discoveries will be 50x faster now that systems can explore huge solution spaces continuously instead of waiting for isolated human insight 👀

The world is about to accelerate beyond imagination. A multi agent system for automating scientific discovery is here. AI Co-Scientists is here. Both papers published 2 days ago. Two huge steps into makig scientific progress 100x - 1000x faster than today.

@EXM7777 A lot of people still think AI increases productivity by replacing work, but it really amplifies the quality of your reasoning for using LLMs/ agents.

the more agents get memory and knowledge bases, the less time i spend on the computer... most of my "work" now is voice notes in telegram... articulating the idea, scoping the task, handing it to an agent that already knows every codex project i have the weird part: i think more than i ever did execution used to eat 90% of the day, now it eats maybe 20 and the bottleneck moved exactly where most people aren't ready for it... the quality of your thinking if you can't articulate what you want with precision, the agent gives you garbage at 10x speed the new skill isn't doing the work, it's knowing what the work even is

@AlexanderKalian Healthy skepticism is fair, but even heavily scaffolded systems solving difficult math problems would still be a meaningful capability jump.

OpenAI claims that an unreleased "internal model" solved a major problem in mathematics. Sounds like OpenAI's very own pre-IPO Mythos moment of overhyping. We should all be asking healthily sceptical questions about this "autonomous" breakthrough: How autonomous was it really? How much scaffolding or chain-of-thought design came from in-house mathematicians targeting this specific problem? How much training data or RAG-based vector database stuff was internally produced data targeted towards this specific problem? How many failed attempts did it make? How was the outcome verified, and how much time and resources did it take, against presumed other failed attempts? They are not gonna tell you - and conveniently, the model is not available for public scrutiny either. AI companies have shifted into "source: trust me bro" mode.

@JimDMiller Cryptography might actually feel the impact earlier since a lot of modern security depends heavily on mathematical hardness assumptions.

Suppose AIs become superhuman at math within two years, as far beyond humans in math as they already are in chess. What practical breakthroughs might follow, perhaps room temperature superconductors?

@Suryanshti777 People underestimate how much of engineering is constraint management rather than typing code itself.

Andrej Karpathy just explained the future of software engineering without directly saying it. The best AI engineers are no longer “prompting.” They’re building systems around the agents. Karpathy’s biggest insight wasn’t: “Claude can code.” It was: LLMs become dramatically better when you force them into disciplined workflows. That’s why "CLAUDE.md" files are suddenly everywhere. Not because they’re prompts. Because they behave like an operating system for the agent. Karpathy called out the exact problems with AI coding: - models assume instead of asking - they overengineer simple tasks - they hide confusion - they rewrite unrelated code - they optimize for completion, not correctness So developers started encoding rules directly into the workflow: → Think before coding → Simplicity first → Surgical edits only → Goal-driven execution And the results are wild. People are now running multiple Claude Code agents in parallel like engineering teams: • one agent researching • one debugging • one writing tests • one optimizing code • one validating outputs Not “AI assistance.” Actual orchestration. And this part from Karpathy changes everything: “Don’t tell the model what to do. Give it success criteria and let it loop.” That is the shift. From: “write this function” To: “here’s the goal, constraints, tests, and verification system — now iterate until correct.” The craziest part? This already feels like a phase shift in engineering. A lot of developers quietly went from: 80% manual coding → to 80% agent-driven coding in just months. Not because AI became perfect. Because the leverage became impossible to ignore. We’re entering an era where the highest leverage engineers won’t necessarily be the best coders. They’ll be the people who build the best systems around AI agents.

@daniel_mac8 Huge credit to the researchers honestly because breakthroughs like this are products of years of infra & training 👏

🤯 GPT-6 is going to be WILD. "The proof came from an internal general purpose reasoning model. Not a system trained specifically for mathematics." AI can create new knowledge, for real. This is the intellectual equivalent of splitting the atom. I can't believe we're alive to witness this.

@gdb Historical achievement for AI!

An OpenAI model has achieved a major breakthrough in mathematics, by disproving a central conjecture in discrete geometry that was first posed by Paul Erdős in 1946. This is the first time AI has autonomously solved a prominent open problem central to a field of mathematics.

@OpenAI genuinely one of the most impressive things I've seen come out of AI research 👏 we've come a long way

Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946. For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids. An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better. This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.

@kimmonismus First one to achieve or get close to it defines what is AGI

As excited as I am that a takeoff seems to have begun and we are entering a golden age of science, one thing remains: I'm hearing more and more from all sides that AGI is within reach. This applies to Google (AGI, Physical AGI), as well as OpenAI and Anthropic. The only caveat: there's no unified definition of what AGI actually is. There have been attempts to standardize a definition, and in my opinion, the most sensible one is Google DeepMind's. But as long as we're talking about different things, it's difficult to find common ground to say *when* AGI will be achieved (which AGI).

@sama how close are we to AGI? 👀

three of the things we are most excited about: 1. AGI accelerating research 2. AGI accelerating companies 3. personal AGI accelerating everyone in achieving their goals today it was great to announce the unit distance result. yesterday it was great to announce that we are offering to invest $2M in openai credits into every YC company. now we need to increase our efforts on the third!

@SalsaTekila What's your workflow like to save usage cause that number doesn't add up for most people 🤔

I spent probably around 35hours going back and forth with Claude on various projects and quests this week. And I still got a good chunk of room to continue until tomorrow's reset. How y'all even hit limits at 20x

@emollick Having a solid memory layer across wins over fragmented tools every time.

The gap between what you can do on ChatGPT/Codex and Claude/Code/Cowork is closing, as Anthropic & OpenAI converge on a single experience. Google's experiences are diverging: Studio & Gemini & Antigravity & the other Google AI apps are increasingly different. Which will win?

@OpenAIDevs async coding from your phone is now real? thats amazing 👀

Your laptop can stay home. Work with Codex from the ChatGPT mobile app, answer questions on the go, and pick up the same thread later from your computer.

@mitchellh Most teams update dependencies without reading the commits and are only now realizing how exposed that makes them.

Fork your dependencies, trim them to only your use case, never update unless it breaks for your users. I’ve been vocal about this for 10+ years. I’ve always said that updating is way riskier than latent bugs (which can be tracked and CVEs monitored). If you are updating a dependency, it’s on you to analyze every single commit in the full transitive set of dependencies. If you dont see anything compelling, dont update! I remember at HashiCorp once in awhile an engineer would try to update a dep or replace a DIY lib with an external one and id always ask “show me the commit we need.” Dont update for the sake of it. Feeling pretty swell about this mentality with all the supply chain attacks happening.