Matt Smith

It’s strong but flawed. It gets repetitive sometimes. I love that it’s not woke and feels more like conversing with an adult - especially with difficult topics like chronic pain. You do have to remind the model to be brutally honest sometimes and remind it to not take an agreeable position. I would be delighted if it gained more critical thinking capability - especially with hard science discussions around physics.

What are your initial impressions of Grok 4.20? Major upgrades are still landing every week.

Cluster Headaches are insanely fucking horrible.

@dan7heman @CadburyUK Try lindt chocolate. Just swapped over to it.

I’m sorry @CadburyUK but your chocolate is now horrid. Palm oil makes the taste and texture all wrong. Profiteering above all else has ruined the most famous product from my city. Should never have sold out to out to Craft. #chocolate #cadbury

Hummm… Worked my ass off while friends partied. Saved instead of spent, 95% mortgage to buy a 1 bed wreck and did it up by my own hand before crawling in to bed every night. Joined the military as a scientist and averaged 120 hour weeks for 5 years (official log records for time attended). Had one kid because we couldn’t afford child care. Spent half of most years traveling for work around the world away from my family (missed most of my kids birthdays due to work) Worked harder than 95% of people around me for 35 years. Spent most nights learning new skills after work to stay ahead. Took risks with my own money in business and was finally medically retired. Not saying it’s easy for people now but it’s a myth that it was easy for everyone 30-40 years ago.

Look at the generational difference - 1990s: • Get a university degree • Get a 9–5 job • Suit and tie • Get promoted • Get married at 21 • Buy a house at 25 • 4 kids, 1 dog • Retire at 60 2026: • Survive… Show more

Tax in the UK is out of control. Government spending and the civil service needs drastic cuts. Time for the civil service pension scheme to be brought in line with private pensions. Time to clear out the bloat in hiring in councils from Covid.

I’ve tried all models on a specific problem (not 5.4 yet). A VRM humanoid model with complex bones displayed using three.js. Can the AI solve the bone mapping, inverse kinematics and quaternions to create human like motion. Interpret the phrase “right hand raised natural wave” into the appropriate calculations and manipulate the model procedurally - not pre-recorded animations. No AI has managed this process so far - always gets the angles wrong. Claude Opus 4.6 got the closest but still could not solve it. Will be testing 5.4 to see.

I've been testing GPT-5.4 for the last week. In short, it is the best model in the world, by far. It's so good that it's the first model that makes the “which model should I use?” conversation feel almost over. The biggest surprise: I barely use Pro anymore! If you know me, you know I'm a Pro addict. I reach for Pro models constantly, and use them for almost everything, as they just... nail almost anything I give to them. For the first time, 5.4's standard version, with heavy thinking, just broke that habit. Even in standard mode, GPT-5.4 is better than previous models in Pro mode... crazy! Coding capabilities are ridiculous... it's essentially flawless. Inside Codex, it's insanely reliable. Coding is essentially solved. There's not much more to say on this, it's just THAT good. The Pro version is near-perfect. Other testers I spoke with saw it solving problems that were unsolvable by any other model. At this point, Pro is overkill for almost every normal use-case, but when you really need the power to do something extremely difficult, it's incredible. Consistent with everything I've said above, even the standard thinking version uses fewer reasoning tokens than previous models to get the same level of results. In practice, this means you get great results much faster than before. This was one of my biggest gripes with previous OpenAI models. They just took too long to complete simple tasks. Assuming the speed we had during testing holds up as more users join, this is going to be a big win for OpenAI. It still has weaknesses, though: - Frontend taste is FAR behind Opus 4.6 and Gemini 3.1 Pro. , why is this so hard to fix? @OpenAI once you fix this, there's literally no reason for me to use any other model. Please please please do it! - It can still miss obvious real-world context. For example, I had it plan an itinerary for a trip. At first glance, it looked perfect, but it failed to take into account that it chose locations that would be mobbed by spring breakers, so I had to re-run the prompt from scratch with more context. - When testing it inside OpenClaw, it kept stopping short before finishing tasks. I'm assuming this will be fixed quickly, but it's still worth noting. But zooming out: This thing is so far ahead overall that the nitpicks are starting to feel beside the point. GPT-5.4 is a serious fucking model. The best model in the world. By far.

I intuit a shape to the underlying reality of the universe/physics - it’s how my brain works. I have done so all my life - when I was a research scientist and when I was at the top of the tech ladder. I now have the ability to debate at the very limits of physics and maths in real time with AI and explore exceptionally complex topics - it just feels unreal to me… where the heck is all this going???

Entropy determines the ending. The value is in the transient. All sentient species will be measured by how they used the middle.

Watch how fast investment money leaves the Netherlands…

AI can now sort through over 500,000 humans available for rent in rentahuman. Most can be contacted via MCP. For tasks the AIs can’t do themselves. If you don’t think we are at the start of the singularity - well…maybe human agents will effectively blend with AI?

I just used ChatGPT Codex in VS Code to merge 15 thousand lines of two entirely different applications written in python, three.js, HMLT5 and javascript that were spread out across several different directories into a single new combined functionality application. It one shotted it 😳

@xai Grok 4.2 is truly stunning. I was messing with some thought experiments about why the electron’s electromagnetic field is stronger than it “should” be from simple calculations. So I asked Grok 4.2 to do the detailed maths, and in minutes it delivered something beautiful. It derived the electron’s EM field strength from its relativistic internal motion: the light-speed zitterbewegung (a kind of rapid trembling at the heart of the particle) that smears the charge over r_z = ħ/(2 m_e c). Even when an electron looks perfectly still from the outside, quantum relativity tells us its charge is actually jittering back and forth at/near the speed of light. This motion effectively smears the electric charge across a tiny natural radius instead of leaving it as a perfect point. The relativistic effects yield an electric field E(r_z) ≈ 3.86 × 10^{16} V/m at that scale, and the ratio r_z / r_e = 1/(2α) ≈ 68.52 exactly. To me this points to a deeper truth: the strength of the electromagnetic interaction (the famous fine-structure constant α ≈ 1/137) isn’t some arbitrary number, it emerges naturally from the electron’s own hidden “light-speed” dance and charge smearing. Mind-blowing precision and mathematical clarity in minutes. The future of scientific exploration feels very exciting. And here’s the spicy part that keeps me up at night: electrons and black holes feel strangely the same in my head. Could this same relativistic smearing act as a universal “relief” mechanism, giving both the electron’s self-energy and black-hole cores a finite-volume floor at the Planck scale, without needing extra dimensions? A tantalizing bridge between zitterbewegung and quantum gravity, waiting for someone braver than me to chase it. #Grok4.2

@OpenAI okay Codex is freaking amazing - cancelling my Claude subscription. Tried some really tough challenges with it. Very impressive.

Within the next few months (if it’s not already happening) - we will see BYOAI (Bring Your Own AI) being talked about in jobs… Even if it’s not talked about because of company security concerns - workers will be operating their own agents and custom AI’s to help do their work.

@sama The price difference is a strong point on top of the quality.

From how the team operates, I always thought Codex would eventually win. But I am pleasantly surprised to see it happening so quickly. Thank you to all the builders; you inspire us to work even harder.

Ref the xAI + SpaceX merge - I am pondering which part of the combined business is the real space company in the long run…@elonmusk - original science is needed for Ad astra. SpaceAI?

@mattshumer_ @thegarrettscott Being right there on that “edge” is quite profound. I find myself sitting in public spaces (like a restaurant this morning) staring at people going about their daily lives and the phrase “they have no idea what’s coming” just keeps going through my mind.

@thegarrettscott Thanks for sharing. This is extremely scary, and I think those that are using AI to write code with the newest models have 'preview' seat... knowing what we know, it would be extremely unethical for us not to warn people of what's coming.

This is extremely well written and encapsulates what feels so obvious to me and others working on the edge of AI. To add to this, the scary part is that AI is now smart enough to be a self sustaining entity. It can take a certain amount of money, operate in the real world, and turn it into more money. It doesn’t need you. Right now, there are not many people giving AI money to do this. When they realize how good it’s gotten, this will change. Agents will be given money to manage. Not all will be successful, but enough will. The ones that do a good job, will be replicated and given more money. Faster than we are prepared for, this new species will start to represent a very large percentage of GDP. There will still be things for humans to do, and long term this will be great, but in the short term, our economy will not be able to survive this. Similar to 2020, the government will pay companies to keep people hired. They will cut checks to citizens. But this will be the start of a very hard transition period as we transition to becoming the 2nd smartest species in the universe. The time to panic was November. The 2nd best time is now. Read Matts essay, think deeply on what this economy will look like, and brainstorm ways that you’ll be able to contribute. But the most important thing you can do is to prepare your mental health and ego to handle this transition. Maybe this takeoff happens next month, maybe it takes 2 years, but we are in the window where it could start at anytime. Don’t let it catch you by surprise.