Uccio

@ercwl So true

AI tools have gotten seriously good. It's hard to point to a specific change, as with many things it's an emergent property of many small things coming together. Claude 4.7 can now string together enough tool calls in a series to competently execute long tasks where it would otherwise fail. ChatGPT Pro can repeatedly look over the same document over and over without context length degrading output into a sloppy mush. Dispatch can control your desktop applications. Agent steering is now possible even without using Hermes/OpenClaw. So you can via voice notes tell Claude to control Codex to run a 12 hour coding /goal, which it will do and even summarize the results to you nicely. Codex is great at execution, Claude is better and front-end and visualizing, cleaning up a bulky data harvest project into something presentable. Visualization is incredible: setting up a webpage with pages you swipe between gives you superpowers beyond what a PDF/powerpoint-based pitch deck could do. .JSX or .HTML-based visualizations built-in in Claude are also excellent, as are ChatGPT Images 2.0. The "needle in the haystack"-scores across models (MRCR) have risen from 20-40% to >90% at 250k context length. Still >75% at 1M context length. I think this is the key ingredient to how these models subtly got better. Being able to recall a single sentence within a long conversation is what helps the model course-correct and stay on track. It's what catapulted Claude's METR score by 3x between Opus 4.5 and 4.6, from ~5 hours to ~15 hours (length of human-time tasks it can handle). Compacting features are still rough but get the job done somewhat okay. Less errors. I've found that I'm more likely to hit the Claude Max (5x) session limit before I hit "context length issues" or another fault that breaks the task. When it exceeds its maximum tool call limit, it competently picks up where it left off. Not sure what this all means. But we've reached the stage where these tools are starting to feel "whole" in a way. With scaffolding improvements and RL-training on tool call sequences, LLM capabilities are starting to exceed what I think most people thought was possible or expected. If it's not working for you, you're probably still managing to use these tools wrong (starting to become an impressive achievement on its own tbh).

@tylerxhobbs @shape ( 02 / 10 ) Tyler Hobbs

Please Respond ( 01 / 10 ) Tyler Hobbs Collecting opens today at 11 AM ET on @shape.

@worldlibertyfi Integrity 🤣🤣🤣🤣

We are taking legal action to hold Sun accountable for defamation. We invite the community to read the full complaint below. And despite the drama, we will stay laser-focused on building with integrity and transparency. url.us.m.mimecastprotect.com/s/Xuv2C0RzARiJ…

Today, we are filing a lawsuit against Justin Sun for defamation. Sun has launched a coordinated media smear campaign against World Liberty Financial and refused to stop even when confronted with the truth. Here's the story.🧵

@EJRWEB3 You've caught my interested

Currently on a three month grind to launch what I believe will be the NFT Project of the Year for 2026. The art style is exceptional, intentionally crafted to resonate with collectors while remaining commercially compelling for both Web3 natives and a massive Web2 audience. We have brought together two elite illustrators with credits spanning Rick & Morty, McDonald’s Anime campaign from 2023, Marvel, and Hasbro working alongside a world class lead creative and branding powerhouse with experience across Mercedes-Benz, Ogilvy, American Express, and Razer. On the technical side, we have two of the top developers in the space. The collection introduces an enhanced fully dynamic trait system, pushing the standard forward, and will be the first of its kind deployed on ETH mainnet. We are also launching a custom closet interface that allows seamless trait management along with features like gender swapping. At the core is a true flywheel. Fast moving, attention capturing, and built for longevity through continuous trait events, memetic drops, and collaborations with artists and brands across both Web3 and Web2. This is not starting from zero. It is an extension of an existing brand already generating close to 40 million monthly views, excluding GIPHY, driven by a fully organic audience and high performing animated content. Mint price will be roughly the cost of a large Starbucks coffee. The reason we can offer that is simple. We have already built the infrastructure. Instead of waiting months for momentum, this launches with it.

@TheMorrisLand @RainyOnMonday @Strandjunker Cause they are morons

@RainyOnMonday @Strandjunker NATO is defensive. They do not start wars. Why is that so difficult for so many of you to understand?

US troops are not in Germany to protect Germany. They’re there because the US needs bases to conduct operations beyond the North American continent. If Trump withdraws them, he will harm the United States, not Germany. It’s absurd how many Americans fail to grasp this.

Our first dive into Multi-Agent Coordination and Cooperation is here, with Hermes Agent Kanban Orchestrate tasks across multiple agent profiles and dependencies easily and visually. Achieve more. See the docs here: hermes-agent.nousresearch.com/docs/user-guid…

@Teknium Just tested it and holy shit

@a1lon9 Just give us the airdrop or stfu

today is a turning point for $PUMP and pump fun I want to give more context on the bigger picture and where we're actually going. over the past ~9 months, 100% of revenue went into buybacks. basically no other platform in crypto has done that at this scale. however, we received ongoing feedback specifically on the feeling of a lack of trust - in the certainty of buybacks, in what would happen to the bought-back tokens, even in whether the business itself would be here in a year. today, we’re changing that. it started with burning ~$370M worth of $PUMP purchases. ~36% of the circulating supply removed from circulation, forever. but that isn’t enough. we’ve also allocated 50% of our next year of revenue to programmatic buybacks & burns. no more uncertainty for those who believe in us & those we’re proud to call our community. but why not 100%? the short answer is the business simply needs the other 50% to grow. a large treasury gives us the flexibility to make big bets over the next 5-10 years, and 50% of ongoing revenue enables us to build better products, infrastructure & reinvest into the ecosystem. I am extremely confident that 50% of the business we're building toward will dwarf 100% of the business we have today.

@zandyor 2yqWyisK28TqtAXnLG8z3pXUom9DYbLZCHvHFBA6joDH

Running an experiment. Launching a memecoin on Solana, airdropping to the first 1,000 wallets that comment in the next 24 hours. Curious what price action and holder behavior look like when the top of the distribution isn't dominated by people who farm every launch. Drop your wallet address.

@MilksandMatcha Working with SME in Italy to help them automate all kind of processes. Would be great help since i'm just starting out

Giving away 5 more Codex Pro plans for folks to try out multi-agent workflows with Codex and Codex Spark Each person will get 3 months of free Codex Pro (highest tier). Winners will be selected from comments in 48 hours, comment below why you want it.

@Brucenewsreview @MykhailoRohoza You are as dumb as you look

@MykhailoRohoza Ukraine has Zero to do with Iran and its parasitic regime of IRGC repressive terrorists who shoot innocent civilians who dare stand up to an oppressive regime.

During talks between U.S. and Iranian delegations in Pakistan, the topic of Ukraine unexpectedly came up. While discussing U.S. guarantees in the event that Iran halts uranium enrichment and abandons the development of nuclear weapons, the Iranian side asked how the United States is fulfilling the 1994 Budapest Memorandum, in which it pledged to respect and protect Ukraine’s sovereignty. The U.S. delegation left the question unanswered.

@Shy100x I am brother

Who’s active? Feel like sending some $SOL to a few people

@stokel @thsottiaux everything is broken

@thsottiaux hello - can we get a reset please? Codex broken

@hexyn7 @Teknium Ask hermes what model he's using under the hood for subtasks, it will tell you and then you can just ask him to change them, or from the Yami file I believe

@Uccio_nft @Teknium how do you do？

请教下@Teknium 铁哥，为啥最近hermes消耗token巨快，同情况下claude code能用很久。

@NVIDIAGeForce PRAGMATA RTX

PRAGMATA has launched with #RTXON, featuring path tracing and DLSS 4! To celebrate, we are giving away this custom wrapped GeForce RTX 5090 featuring Hugh and Diana, perfect for the adventure that awaits on the moon. Want it? Comment "PRAGMATA RTX" to enter!

@ic3_nomad @1PotatoDog @hexyn7 @Teknium You can't use oauth but you can give it Claude code with ACP, I'm running it like that and it works great

@1PotatoDog @hexyn7 @Teknium Don't do it. You'll get banned. And still you will get millions of tokens used.

@gkisokay I need to somehow optimize my usage, running through my gpt max plan after 3 days on avg

I guess you won't be burning those Claude credits on your agents any longer, which is fine because there are better options. My setup to run my agents all day long is $30/mo: - Qwen3.5 9b local for brainstorming ideas 24/7: free - Minimax m2.7 for most coding and workflow tasks: ($10/mo, 1500 calls/5hr) - ChatGPT 5.4/mini for the main brain to keep everything in line ($20/mo, generous weekly limits) I hope more people will wake up to experimenting with more models. For example, Minimax M2.7 can handle 99% of your work for such a great price. Claude has been instrumental to getting us to where we are today, but it is not necessary any longer.

@sudoingX I'm considering getting my first local set up, I have studied and have a pretty good idea on what to get. Can I show you and hear your opinion?

the people telling you a single 3090 can't ship production quality are not wrong about the ceiling. they're wrong about the conclusion. most of them prompted a model twice, watched it hallucinate, and made a youtube video titled "local AI is NOT ready." they never iterated. never pushed context. never matched the right model to the right task. never configured a single flag. they gave up where the work starts. and i never said you'd run frontier agents on a 3090. go back through everything i've posted. what i said is you start there. you learn what your workload actually needs before someone else defines it for you. because the ones telling you local can't compete have API subscriptions to sell. they have wrappers to maintain. they have inference margins that disappear the moment you run your own hardware. they are not protecting you from frustration. they are protecting their revenue from your independence. every local AI transition destroys someone's SaaS margin. that's why they fight it. buy a single GPU. ROME or EPYC board, ECC memory, scalable from day one. cancel every subscription. run local until you hit a wall. and when you hit that wall you will know exactly what cloud compute you actually need instead of what someone told you to buy. use cloud to mine its intelligence then bring it back to your hardware. and scale your GPUs as you go. that's the whole point. don't build on someone else's thinking. don't store your reasoning on someone else's servers. own the hardware. own the frustration. own what comes after it. this is what i've been saying since my first post on this account and it's what i will keep saying. there is nothing you can do about it.