Fei Li

@sama GPT-5.4 feels like it just learned to walk and 5.5 is already running. At this rate, we’ll be at version 6.0 before I’ve even finished reading the documentation for the last API update. The pace is exhilarating but also slightly terrifying.

GPT-5.5 is here! We hope it's useful to you. I personally like it.

@GadzhiIman Top-line revenue is a vanity metric; liquid cash flow is a sanity metric. Most 'big' businesses are just glorified cash-recycling machines. Info is one of the few models where you actually get to keep the money you make.

The info business is the best cash flow business on earth. I've seen every business model up close and nothing comes close because they all rely on the lie of top line revenue. You'll see businesses doing $30M a year and think they're killing it, but then you look at the profit and 60-90% of it has to be reinvested just to keep growing. You can’t take any money out of the business. Info is the complete opposite because you don’t need to rely on an arbitrary networth. You get to have actual power to make moves because you have liquid cash.

@rajshamani True, but only for the first transaction. Distribution gets you the meeting, but product quality determines if they ever come back. Winning isn't just about reaching the audience; it's about not burning their trust once you have it.

Your ability to sell what you make is worth more than what you make. An average product with distribution beats a great product with no audience.

Mastering AI Agents actually requires a specific skill set. It’s not "zero barrier." Even basic prompt engineering is a huge hurdle for most people. In traditional industries like real estate and manufacturing, the learning curve is incredibly steep. The best path forward is building and deploying ready-to-use agent systems directly into their workflows. E-commerce is slightly ahead because they’re used to SaaS platforms, but traditional sectors are still struggling. Right now, the people actually paying for AI are tech-heavy early adopters. It’s rare to see widespread use. The exception is AI-focused companies that live and breathe agentic workflows. Their efficiency is insane. We’re talking about 10-person teams doing the work of 100 people. The productivity gap is widening, but many founders don't realize it yet. If you aren't hands-on with AI as a founder, you can't see the "new" ceiling for your team's capability. You won't know what can be offloaded to AI, which makes scaling efficiency a nightmare. Most users drop off after 1 or 2 tries because the output didn't "wow" them. In many cases, the agent isn't the problem; the user just hasn't mastered how to steer it. We still have a long way to go in making these tools truly intuitive. At this point, I talk to AI more than I talk to humans. Why? Because LLMs are smarter than the vast majority of people out there. They’re definitely smarter than I am!

@kunalb11 Agreed. AI makes a lack of focus look like high-speed iteration. If you can juggle 5 problems with the same mental energy it used to take for 1, then sticking to 'just one thing' is effectively leaving leverage on the table.

Pre AI for tech companies focus was key. Post AI wonder if focus become a limiting factor.

@yasser_elsaid_ AI promised us the 4-hour workweek, but instead, it gave us the ability to do 80 hours of work in 40. We’re just running faster on a much more powerful treadmill.

AI promised to do the work for us so we could enjoy our time doing other things. Since llms, me and everyone ambitious around me has been working harder than ever. I don't think this stops anytime soon.

Turning into an AI influencer was never the plan, but here we are.😂 I’m getting tons of requests from AI Agent startups for paid promos, but most of their products are just unfinished MVPs. I see other creators praising everything for money, but that’s not my style. Recommending a tool that doesn’t work is the fastest way to lose credibility. We only promote products that solve real problems. Period. ⚡️

@thsottiaux Don't build a world-class dev platform by building walls. Keeping the barrier to entry low is the only way to stay the industry standard.

I don't know what they are doing over there, but Codex will continue to be available both in the FREE and PLUS ($20) plans. We have the compute and efficient models to support it. For important changes, we will engage with the community well ahead of making them. Transparency and trust are two principles we will not break, even if it means momentarily earning less. A reminder that you vote with your subscription for the values you want to see in this world.

@thejustinwelsh Feeling stupid is just the brain's way of re-mapping. The 1% aren't smarter, they just have a higher bandwidth for temporary embarrassment.

If you can spend a decade showing up, feeling stupid, getting feedback, trying again, feeling stupid again, and eventually getting better, you'll likely be in the top 1% of worldwide earners. The path is always the same, but few people will actually walk it.

Saw good news about Cursor today—genuinely happy. Great products + great teams deserve a bigger stage. But let’s be real: with Claude Code rising fast, Cursor (without its own model) has been under serious pressure. As a heavy user, I’ve been worried it might fade. Cursor actually changed things for me. No engineering background, but I was able to tweak frontend code and build demos early on. It made tech feel approachable—and even fun. xAI’s LLM is pretty rough right now, but their data advantage is real. Wouldn’t be surprised if they catch up fast. The ecosystem needs more players. More competition = more choices for users. Also—hot take: I’ve moved from GPT to Gemini. GPT’s error rate just feels higher lately. And for investing / auto-trading—some models from Chinese companies are surprisingly strong. In some cases, even outperforming top U.S. models. Worth exploring. @cursor_ai #llm #investment #ChineseLanguageDay #agi

@paulg Exactly. In a world of simultaneous discovery, the only moat left is the speed of iteration.

Interesting ideas tend to appear in multiple places in slightly different forms. Though that's somewhat of a tautology, because this is a big part of what makes them interesting.

@OpenAI If startups and indie devs are excluded from the 'legitimate defenders' tier, the ecosystem resilience OpenAI is talking about will remain a walled garden.

For years, we’ve been building our cyber defense program on the principles of democratized access, iterative deployment, and ecosystem resilience. As model capabilities advance, our approach is to scale cyber defense in lockstep: broadening access for legitimate defenders while continuing to strengthen safeguards.

We’re expanding Trusted Access for Cyber with additional tiers for authenticated cybersecurity defenders. Customers in the highest tiers can request access to GPT-5.4-Cyber, a version of GPT-5.4 fine-tuned for cybersecurity use cases, enabling more advanced defensive workflows. openai.com/index/scaling-…

Founders and Solopreneurs shouldn’t have to struggle with manual coding or high design costs. This all-in-one AI Agent manages your full product launch pipeline, allowing you to go from a single prompt to a high-fidelity website or app almost instantly. With one click, this AI Agent transforms your vision into a pixel-perfect UI. Whether you’re a startup founder shipping an MVP without a CTO, a developer who lacks design aesthetics awareness, or a PM who needs to quickly obtain , this agent handles the heavy lifting. Go from a rough idea to a live, production-ready product in seconds. All with zero code. 👉Try it for free: agentum.me/marketplace?ag… Meet Agentum Tell us what you’re trying to get done. We’ll recommend the right Agent fast so you can work in an AI native way and boost productivity by 10x+. One platform One Workflow Drive your business growth automatically At Agentum, our vision is simple yet powerful: to make AI agents accessible to every business. We believe that every business, regardless of size, deserves the opportunity to harness the incredible power of AI to streamline operations, boost productivity, and unlock new growth. #Agentum #AIAgents #AINative #NoCode #Productivity #BusinessGrowth #AIWorkflow #WebDesign #SaaS #Startup #Automation #BuildInPublic #MVP

@SahilBloom Spot on. 90% of the competition drops out during the first 20 boring workouts. Consistency isn't just a habit; it's a competitive advantage in a world addicted to instant gratification.

Every single thing you want in life is on the other side of something that sucks. That suck might be 100 workouts, 100 bland meals, 100 hours of work, or 100 hard conversations. Embrace it as the cost of entry. The answers you seek are found in the actions you avoid.

@Dwriteway Facts. I’ve seen 80% of marketing fail simply because they led with 'how it works' instead of 'why it hurts.' Winning the argument of the problem is 90% of winning the sale.

The gurus won’t tell you this; Never lead with your product. Lead with your customer's problem. Describe it better than they can. Then gently reveal that you have the answer. Makes things 100× easier.

@Dwriteway Spot on. The 'hidden cost' of cheap clients isn't just the money. It’s the opportunity cost. Every hour spent on a low-budget revision is an hour you can't spend on a high-value partnership.

Cheap clients are the most expensive ones you'll ever have. The less they pay the more they complain.

@pmitu Facts. The cost of building a feature nobody wants is 100x higher than the cost of a landing page that fails to convert. Sell the 'why' before you spend a single cent on the 'how'.

Marketing starts right after you have new idea. If you wait until the code is finished, you’ve already lost.

@AnCapibara @unclebobmartin That just moves the bias from the 'training data' to the another thing. A truly unbiased experiment requires the AI to find the signal in the noise, not just echo our fine-tuning instructions. Still a human-in-the-loop problem at the end of the day.

@box6_box @unclebobmartin What if the fine-tuning post-training includes this kind of experiment?

One of the interesting things about AI development is that we can now do experiments of process and technique without a significant human bias. Is dynamic typing better than static typing? Are short iterations better than long iterations? We can do those experiments now.

@trikcode 11% cheaper is just the baseline, the real value is in the optimized throughput.

Anthropic's new strategy is smart use Opus to think use Haiku to execute 11% cheaper and actually performs better