@exm7777Oct 27

let me tell you something my friend this is not the time to cancel subscriptions this is not the time to play it safe this is not the time to "wait and see" right now is the time to go fucking ALL IN >max out your Claude sub >burn through your ChatGPT plan >spend all your tokens on OpenRouter and top up AGAIN it's time to experiment, to build and PRINT a shit ton of money

@alliekmillerOct 27

I don't think people realize how many pitches investors get today from people who "invented a new algorithm" or "solved the climate crisis" because of one conversation with ChatGPT.

@pmituOct 27

POWERPOINT JUST GOT ITS BIGGEST AI UPGRADE everyone on X is building with Claude and ChatGPT meanwhile, millions of businesses still run on PowerPoint Templafy just launched Document Agents: → creates, edits, and reviews presentations → knows your brand and company rules → works across PowerPoint, Claude, ChatGPT, and Copilot → cuts deck creation from 2 hours to 8 minutes, on average AI helps us create more stuff decreasing hours of boring work into minutes

@aakashguptaOct 27

Sam Altman asked ChatGPT why his baby keeps throwing pizza on the floor and laughing. Scientists at Johns Hopkins answered that 11 years ago, and honestly the answer is a little embarrassing for the guy who runs ChatGPT. If your kid ever did this to you at dinner, you were watching a physics experiment. Two researchers, Aimee Stahl and Lisa Feigenson, brought 110 babies into their lab, all around 11 months old, and showed them toys doing impossible stuff. Some of the babies watched a ball roll right through what looked like a solid wall. Others watched a toy car roll off a ledge and just hang there in the air. Then they handed the babies the toy. The wall babies grabbed it and banged it on the table, like they were checking whether it was actually solid. The floating-car babies kept dropping theirs on the floor. Each group went for whichever test matched the rule they'd just seen broken, and the babies who watched the toy act normal barely did either one. None of these kids could talk yet, and nobody had ever explained gravity to them. They still showed up with a pretty good idea of how stuff is supposed to behave, and the moment the world didn't match it, they went and got more data. So a baby dropping the same slice of pizza over and over is basically rerunning the experiment to make sure food still falls. Around that age they also start watching your face to figure out how they should feel about what just happened, which means your reaction is part of the experiment too. Here's where it gets uncomfortable for AI. Yann LeCun, one of the godfathers of modern AI, did the math on this. A 4-year-old has been awake for about 16,000 hours, and just through their eyes they've taken in roughly as much raw data as the biggest AI models get trained on. The kid's data comes as live video, with hands to poke things and a parent reacting to every move. LeCun thinks that's what today's AI is missing, and it's a big part of why he left Meta to build something different. So the guy running OpenAI asked a chatbot to explain the one learning machine nobody in AI knows how to build, and it's been sitting in a high chair in his kitchen this whole time.

@mtsliveOct 27

a16z's @droovg explains why ChatGPT and Gemini struggle to find the best startups to work for, and how Cosign is building the missing data layer: "One thing that LinkedIn will never do is introduce an MCP that allows you to have vast search access over the graph." "When I was in New York this summer, I wanted to start meeting people who I thought I might be able to work for one day." "So I Googled, who are the best startups in New York? Gemini gave me a pretty bad answer. I switched to ChatGPT, it also gave me a pretty bad answer. That whole data layer of information is kind of missing right now." "We have an iMessage agent. We also have a personal MCP that lets you connect your Muse, your Grok Bot, your Instinct to the platform, to make this truly the first agent-native professional networking platform." @cosign @a16z

@businessbaristaOct 26

Experian, a $35 billion business with roots dating back to 1826, is surprisingly AI-native. Its new Chief AI Officer, Vijay Mehta, broke down the credit bureau’s end-to-end AI strategy. Complete list of notes... Big takeaways:
 • Experian is well past proof of concept, with AI capabilities running in production throughout the company.
 • The next stage is what Mehta calls "reimagination." Redesigning how work gets done through delegation to agents vs. prompting chat bots. 
 • Adoption is close to universal. 85% of employees actively use the company's AI tools, 88% of developers use coding assistants, and employees built more than 6,000 custom GPTs by the first half of fiscal 2026.
 • They're seeing real ROI. It's measured in time savings, time to value, quality, velocity and traditional before and after financial metrics, not in adoption or token use. How they built the foundation:
 • Within months of ChatGPT's launch, they trained thousands of employees and put AI policies and guardrails in place.
 • AI traffic was routed through a control plane that gives them observability, monitoring and security.
 • They wrote a GenAI risk framework and built the Experian GenAI Platform, an internal system that lets employees build internal-facing & and customer-facing apps, while accessing company data securely.
 • They rolled out a coding assistant to their large developer base and built their own Experian GPT into products like a digital financial assistant and their Ascend platform.
 • They created a GenAI Center of Expertise, which now sits under Mehta's AI office.
 • Since then, AI has spread across software development, analytics and business operations. AI inside the employee experience:
 • MyCareer platform recommends learning and career moves and helps managers spot skills gaps.
 • Nadia, an internal AI coach, helps employees set goals, build development plans and prepare for performance reviews.
 • Nadia also gives leaders personalized coaching and 360-degree feedback. On models and cost:
 • "Not everybody needs the most advanced frontier model. Most workloads can be done with a free or very low-cost model."
 • He compares reaching for a frontier model by default to using a sledgehammer on a job that needs a much smaller tool.
 • They run a mix of open-weight models adapted to their own environment and frontier models accessed through AWS Bedrock.
 • The control plane routes each workload to a model based on the task, language, region and regulatory requirements. On agent safety:
 • Guardrails are designed into the architecture from the beginning instead of being added later.
 • Their Agent Operating System runs on that control plane, so it can sandbox, monitor and stop an agent process when necessary. "The ability to kill a process, the ability to kill a process, I will repeat that, we've got all of that built in."
 • Their risk profile differs from the frontier labs' because their systems work over a defined body of financial and identity data, without broad access to the internet. "We start with our corpus of data and we put controls around that. And we know exactly how the data is moving." What the CAIO role actually is:
 • Experian is a big company split into many business units. His job is one common AI strategy and direction: "to make sure that we are all marching in the same direction."
 • He accelerates the AI agenda both internally and externally.
 • Risk, security, compliance and partner choices all come together in the AI office. AI as the new middleman:
 • Consumers increasingly start their financial journey inside an AI assistant instead of going to Experian or a bank, so Experian is meeting them there.
 • It's the first personal finance app integrated with Google's Gemini. US consumers who connect their Experian accounts can ask Gemini about their FICO credit rating and what's affecting it.
 • Inside ChatGPT it has launched personal loan shopping, credit card shopping and an insurance marketplace.
 • On the enterprise side, ServiceNow is the first integration partner for its Agent Operating System. Under a multiyear partnership, ServiceNow agents tap Experian's Ascend data and decisioning from inside enterprise workflows.
 • It also launched Agent Trust, which verifies that an AI agent is actually connected to the consumer it claims to represent.
 • Could agents eventually skip Experian and get the data somewhere else? He says they can't reproduce it. Decades of longitudinal financial and identity data, plus its provenance and regulatory standing, are what matter when the output drives consequential decisions. "In order to be able to make a decision, you've got to have that trusted data source." His advice if you're early:
 • Train employees.
 • Set guardrails before deployment. "So much of safety and responsibility comes from the person interacting with the system."
 • Route AI traffic through a controlled gateway where usage can be observed and policies enforced. Which AI leader should I learn from next?

@levelsioOct 26

Google is down bad for no apparent reason So today I have been loading up on $GOOG stock (well $GOOGL but they're mostly the same) for in my eyes very cheap I just Facetime'd with my dad and we talked about ChatGPT and he answered "yes I cancelled my ChatGPT because it's on Google now" I asked what the jackfriks he meant He said "so Google Search has AI now and I trust it more because it links back to sources etc" My survival thesis is we have to be co-owner of BigTech and BigAI while they become to sole owners of exponential intelligence which will replace a lot of our businesses and which they will then sell back to us I'm doing what I'm saying

@chatgptOct 26

Known affiliation

GPT-6.1 Terra means business. Our new fast, efficient model pairs sharper visual reasoning and precise computer use with clearer, more natural writing. From decoding technical drawings to drafting proposals, filling forms to editing videos—it’s built for the whole workday. GPT-6.1 Terra is rolling out today to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex.

@tibo_makerOct 26

there are 2 internets right now 1. where agent swarms run entire companies, everyone’s hiring “AI employees”, and everyone’s chasing the one-person billion dollar company 2. where a real users, paying real money, takes a photo of their laptop screen with their phone to share something the hype wave we all live in makes it feel like everything has changed, and in some ways it has ChatGPT just hit 1.2 billion weekly users, which sounds massive until you realize about 85% of the world still doesn’t use it we’re debating superintelligence while most people are still figuring out how to share a screenshot both of these realities are true at the same time and there’s real money in building for both

@awilkinsonOct 25

Claude and ChatGPT plugins are the death of software. 1. Integrate your software into their interface 2. Get everyone used to using it in that way (because it’s a better experience). 3. Eventually cut you out of the equation. CapCut doesn’t need to exist 😂 It’s an interface for something that doesn’t need an interface anymore. Lambs to the slaughter. They are going to eat effectively all software this way. And APIs aren’t going to save everyone. APIs are just software that computers use.

@rhyssullivanOct 25

30 year prediction that there’s going to be the equivalent of McMansions but from when people send their space to ChatGPT to redesign it

@amix3kOct 25

ChatGPT and Claude aren’t the endgame of AI. Even something like Grok Bot and Muse are early prototypes of the future. I see four levels: 1. Smart features (Google Maps, Todoist Ramble): AI/ML improves part of the experience, e.g., finding a route or capturing tasks. 2. Chat in the app (Copilot): A sidebar with AI helps you work with what’s in the app. 3. Chat is the interface (ChatGPT, Grok Bot): You describe what you want, and the assistant uses tools to do the work. The chat window is still where you go to delegate and follow up. 4. Agents as collaborators (Shopify River, Doistbot): Agents are participants in your work. They use context, ask questions when needed, and deliver results alongside your teammates in *your* tools/environment. Very few individuals and organizations operate at Level 4, but this is the endgame. You can see early glimpses of this, and it’s only in the last few months that Level 4 is even possible.

@aakashguptaOct 24

Is the product builder role real or is it hype? Everyone has a take. Ankit Shukla went and checked. Not his opinion. Not what gets claimed on LinkedIn. He ran 12,500 actual PM job postings from top companies through Claude and GPT to see what employers are writing down when they're trying to hire someone. More than 30% of PM jobs now ask for AI skills. The interesting part is what they ask for inside that. Not only the tools. They want you to know which problems should be solved with AI and which should not. That distinction is the whole thing, and it's the part most people skip while they're learning their fourth automation platform. Knowing n8n is a thing you can acquire in a weekend. Knowing this particular workflow should stay deterministic, or this feature doesn't need a model at all, is judgment that takes exposure to real decisions. The economics follow from what got cheap. Engineering is the part AI already handles. Everything downstream of a good decision collapsed in cost, which makes the decision itself the expensive input. So the premium isn't paying for your tooling. It's paying for your aim. Which is also why this reads as an opening rather than a threat. The scarce skill is the one PMs were already supposed to have, now with a much larger set of things you can actually go build. The title hasn't changed at most companies. The job posting already has.

@a16zOct 23

AI demand is a barbell, with the labs on one end and enterprise cost-cutting on the other. Amazon CEO Andy Jassy says the biggest market sits in the middle, and inference is still "an order of magnitude too expensive" to reach it. "On one end of the barbell are the AI labs who are consuming gobs and gobs of compute, unlike anything I've seen. And then a few runaway apps, ChatGPT, Claude Code, things like that." "On the other end of the barbell are enterprises who are largely using AI for cost avoidance and productivity, so automating customer service or fraud or business process automation." "In the middle of the barbell are the current production workloads of enterprises, many of which are using AWS and using inference today, but I would say the overwhelming majority of which still aren't in any pervasive way." "And that I think is gonna be the largest absolute segment of AI usage." "The cost of inference has come down quite a bit in the last few years, but it's still an order of magnitude too expensive, in my opinion, for it to be used as pervasively as possible." "If you're asking enterprises to rewrite their applications that use CPU today with AI accelerators... the amount of incremental benefit they have to get is a much higher bar to want to rewrite those apps. But if you can take the cost of inference down, that bar is much lower, and all applications are gonna be better and smarter and more capable." @ajassy

@rileybrownOct 23

Every added layer of complexity in an application makes it 10x harder to describe and make content on. Back in April I was making many videos on Codex. "Codex, OpenAI's version of Claude Code" "Codex, the desktop app on your computer" "Codex, the super app that can do any CPU task" Very, very easy to describe. Repeatable, and viral. This made it effortless to describe on short form or long from videos. Then ChatGPT and Codex merged. Okay, now Codex is not an app. Codex is a feature inside ChatGPT. "Codex, the AI coding agent feature inside the ChatGPT Desktop app" Okay, now ChatGPT has "GPT Work" "Codex, the AI coding agent feature inside the ChatGPT desktop app that allows you to do coding tasks, and sometimes work tasks, some of those should be done inside GPT Work, which is kind of similar, but like different, depending on if you are using Cloud GPT Work, or local GPT work, in which case is alot closer to Codex, but still not quite the same. If you try and use the skills you created on Codex they won't work when doing a cloud GPT Work session but if it's a local session it will work." We haven't even made it to Codex Cloud sessions, GPT Dots, and GPT Plugins yet. This goes for anyone building software. If you want to create an ecosystem of creators who talk about your product organically, you should aim for simplicity. BTW not saying it was the wrong decision for OpenAI to merge them, i'm just saying for me it got much harder to describe with every new feature added.

@exm7777Oct 23

there's a simple way to 100x your AI outputs and most people will never do it... agents like Claude Code can spawn sub-agents that run tasks at the same time most people use this for basic stuff the real play is having 3-5 sub-agents review Claude's work from completely different angles... using ANOTHER model (ChatGPT, Grok or anything really) but don't use fake "expert personas" you invented in a prompt, you need real frameworks from real people how to build these expert profiles: option 1: use someone so well-known their thinking is already in the training data > have a Karpathy-style agent tear your systems apart > build a profile based on Ogilvy's principles to roast your copywriting option 2: go deeper with Obsidian and Hermes / Grok bot > feed it an entire youtube channel from someone you trust > or their blog, their newsletter, their podcast transcripts > prompt it to extract an expert card with their frameworks, decision patterns, and principles now upload that material into your agent what happens next: Claude generates the first draft > sub-agent 1 reviews it through expert A's lens > sub-agent 2 stress-tests it with expert B's framework > sub-agent 3 catches what the others missed > all running at the same time Claude thinks through every piece of feedback and rebuilds you only see the final version... the one that survived 3-5 rounds of real scrutiny this is close from getting the absolute best output for a specific task

@itsolelehmannOct 23

if you offer any kind of service you need to level the fuck up or you are toast i'm looking for a new editor right now and got the first test video from an agency and it's literally worse than a one shot prompt with chatgpt if you're only as good or even worse than a one shot, it's over spend as much time getting as cracked as you can good enough is bad now you need to be GREAT not saying this to doom, but so many people are clearly not getting the message that the bar is rising atm

@aakashguptaOct 23

Gemini got asked for a random number between 1 and 10 a hundred times in a row. It said 7 all hundred times. And 7 is also the number you were most likely to pick. That's the entire secret behind AI "reading your mind." You and the chatbot learned what random looks like from the same place, which is other people. People are hilariously bad at being random. In one study of 558 people, 28% picked 7, way more than any other number. GPT said 7 in 92 out of 100 tries and Claude said it 90 times. Watch what your brain does in the half second before you answer. 1 and 10 feel too obvious, so you skip them. 5 feels like a cop-out because it's right in the middle. Even numbers feel too neat, so 2, 4, 6 and 8 are out. 3 and 9 feel patterned, so those go too. Count what's left and you've got 7. An AI learned to talk by reading billions of pages that we wrote, including every forum thread and party trick where somebody typed "pick a number." Inside the model, the only thing happening is next-word prediction. It looks at the question and goes with whatever word most often came next when real people wrote it, and across all those pages, that word is 7. And it exaggerates us. About a quarter of people go with 7. The model's whole job is landing on the single most likely answer, so a quarter of people turns into almost every single response. Same thing happens with bigger ranges. A researcher at Capco asked seven different AI apps for a number between 1 and 50, and six of them said 27. Grok was the holdout with 42, the Hitchhiker's Guide joke. Veritasium surveyed 200,000 people on 1 to 100 and the favorites were 37 and 73, which is exactly where the AIs cluster too. So try it yourself with 4 in your head. A real mind reader would catch that one. ChatGPT is only ever going to catch the person who picked 7, and that person has been typing 7 on the internet for thirty years. Every time an AI "reads your mind," you're looking at the average of millions of people who each felt sure they were being original.

@aakashguptaOct 23

Apple's new folding iPhone costs $1,999 and opens into a screen almost as big as an iPad mini. The first app I've seen that actually makes that inside screen worth the money is a chatbot. Folding phones have been around since 2019, and close to 100 million have shipped. The knock on them the whole time has been that most apps just take the regular phone version and blow it up. You paid an extra thousand bucks for a bigger Instagram. Now look at what's happening in this video. On the left you're texting Claude "Can we make it Friday?" On the right is the actual Saturday coffee-and-pastries poster you're asking about, and Claude is republishing it right there while the chat stays open next to it. AI apps have quietly had two halves for a couple years now. On a laptop, Claude has shown the thing it's building in a side panel since mid-2024, ChatGPT added the same idea with Canvas a few months later, and coding tools like Cursor work that way too. You type on one side and watch it get built on the other. A normal phone can only show you one of those halves at a time. So you send the message, swipe over to check what changed, swipe back to ask for another tweak, and do that dance every single round. That's also why the split screen on folding phones never really caught on. Splitting the screen meant running two different apps at once, and pretty much nobody wants Gmail sitting next to Google Maps on their phone. It looks great in the commercial and then nobody actually does it on the subway. With an AI app, one request fills both sides. The chat and the result belong to the same task, so using the whole screen happens without you thinking about it. Claude Code makes it even more obvious. You've got an agent writing code on one half and the app it's building running live on the other, which is how most developers already work at their desk with two monitors. Now that setup fits in your pocket. Apple will hand an "optimized for iPhone Duo" badge to basically any app that gets recompiled with the latest Xcode. Getting the badge is easy, but deciding what belongs on each half of the screen takes real design work, and chat apps showed up with that answer already built in. Preorders sold out fast enough that new orders won't ship until late January, and Counterpoint thinks Apple moves up to 6 million of these this year. That's a few million people about to carry a screen that's laid out like an AI workspace whether they planned on it or not. Seven years of folding phones, and the thing that finally fills the second screen is whatever you just asked for on the first one.

@aakashguptaOct 23

Wade Foster will tell you not to vibe code your CRM. He is firmly in that camp. He also vibe coded a CRM. The exception he made for himself is the interesting part, because of what he deliberately left out. It is not HubSpot. It is not Salesforce. It does not try to be. No pipeline views, no reporting module, no dashboard anyone logs into. It handles a narrow set of workflows he personally cares about and nothing else. It runs Saturday night. Sunday morning he has 10 to 20 drafted emails waiting, each aimed at an enterprise customer he wants to stay close to, each with a reason attached. A renewal is coming up. Usage has dropped. He reads them, says yes, and fires them off. Then the line that reframes the whole category: he barely opens the thing. Think about what that says about the software it replaced. A CRM is mostly a place to go and look at data. You open it, you scan, you form a view, you decide who to contact. Every one of those steps is work you do because the tool cannot do it for you. He inverted the direction. The system comes to him with the recommendation already made, and his only job is the approval. That is why the don't-vibe-code-your-CRM advice and this project do not actually conflict. Rebuilding Salesforce is insane. Building the four workflows you'd use Salesforce for, with the browsing step deleted, is a weekend. Most enterprise software is a waiting room. You go there to find out what to do. The version worth building tells you.