.

@NickADobos yup... switching auths and providers everyday lol

LLM usage limits being 5 hour windows is a fun way to ensure I swap between codex and Claude every day and have no loyalty to either

@StockSavvyShay frontier labs are eating everything. no one is safe.

Anthropic is reportedly preparing to launch Claude Opus 4.7 and a new AI design tool as soon as this week. The tool is said to create presentations, websites, landing pages and products from natural language prompts which helps explain the pressure on $FIG and $ADBE today.

@claudeai funny how the labs are all converging back to IDE/GUI environments... great form factor, looks polished, but i think there is much more room for improvement in this regard, and this goes for both claude and codex. interested seeing how this space evolves over the coming months

We've redesigned Claude Code on desktop. You can now run multiple Claude sessions side by side from one window, with a new sidebar to manage them all.

if using the 1M context window quality degrades to a noticeable degree around the 350k mark.... but yea codex with the 5.x models handles context so well that i frequently do not have to do compaction unless sessions are too long, and even then, the compaction is the best in the game by a long shot! makes working with codex so much smoother, small things like that makes a world of difference

Yes. With exception to GPT 5.x models. I see no degradation through large context or compaction. It's absolutely insane. The biggest QoL update for me. But I don't use the 1M context window. Context window anxiety begone!

@jullerino waitttt i thought i was the only one with that issue like a hour ago LOL

OpenAI compaction so good they had to take it down 😭😭😭

although that could be the case in general, thats not whats happening here; Baltimore's new DA Ivan Bates is actually doing his job unlike his predecessor Marilyn Mosby. crime is still rampant however, particularly on "petty crime" and nuisances, but homicide rates are objectively improving due to repeat offenders actually staying locked up and not hastily released.

When black mayors take over cities, they stop reporting crime and pressing charges on blacks which gives the appearance of lower crime. The crime increases but off record.

@NousResearch perfect timing, will tune in

Tomorrow: Hermes Agent Jam. Nous Research team, presentations, Q&A. Tuesday April 14th, 4PM EST in the Nous Discord

@iruletheworldmo seems like the time has come again

tomorrow will be magical. big plans. huge.

@chetaslua @mweinbach going to have great fun using this upon release :)

Codex conversion to Super App This is the first look , you can also toggle this looks in your codex , just ask codex to patch codex app OpenAI is cooking very hard this time Credit @mweinbach

@gabrielchua incredible resource, thanks alot for compiling these use cases, i know a few that would benefit from this

Here’s how we use Codex to: > understand large codebases > review PRs faster > build macOS apps > turn Figma into code > automate bug triage > create a CLI as agent tools > analyze datasets > generate slide decks > coordinate new-hire onboarding > learn a new concept …and more. developers.openai.com/codex/use-cases @kagigz , @Dimillian , @nickbaumann_ and team pulled together a great collection of Codex use cases, and based on how engineering *AND* non-engineering teams build with Codex daily at OpenAI. What else should we cover?

@elonmusk human ingenuity at its finest

“They are smart … for humans” – future AI

@NousResearch @JimsYoung_ i still remember those openhermes and atropos days lol you guys came a long way! great seeing the progress and evolution

@JimsYoung_ That is not at all how we made our name lol

Hermes finally hit it big, so I want to write about sth i know about @NousResearch — a truly unusual organization: - Nous Research’s $70M funding round has nothing to do with decentralization or any of that. The real story: someone formerly at Nous once wrote a long prompt that, when sent to Claude, would make Claude go off the rails. Anthropic scrambled to patch it, and then Nous came up with a new prompt… at which point Anthropic basically threw up its hands and resolved the issue through subsequent model updates instead. That’s how Nous Research made its name in one shot. - Nous is genuinely strong at prompting and post-training, with a real track record. When @anyway_sh was just get started, nous people also helped us at prompt eng. - Web3 or not, crypto or not, Nous is an AI lab with great taste and an extremely high density of talent — that’s just the team’s inherent strength. The reality is that crypto is a mixed bag, but it’s also true that there’s real talent in there: the founder of OpenRouter, the founding team of ElevenLabs, etc. - Also, Nous genuinely loves the Chinese internet ecosystem — they’re already on WeChat.

@TheGameandFrame @RedMedia_us people are upset in the replies but this is obviously true... the left have a nasty habit of eating their own often.

@RedMedia_us That’s the problem when you align yourself too much with liberal shit. You gotta be accepting of shit you don’t even like . Being a liberal means you gotta sell yourself for what society thinks is acceptable. She put herself in a box. Can’t even have an opinion FUCK THAT

Sabrina Carpenter comes under fire for saying an audience member's ululation (zalghoota), a traditional celebratory vocalisation common in Arab and North African cultures, for yodelling and repeatedly calling it "weird" during her Coachella set.

@reach_vb @ShanuMathew93 yea the 2x loss is pretty noticeable but was always expected, and even still, the plus plan is still quite generous all things considered. probably just not wise to use the fast mode anymore now, unfortunately

@ShanuMathew93 to be clear, we haven’t diluted the plan, it’s likely due to 2x usage limits promo ending on Plus subscriptions, we’re always striving to provide the best value for money to our users! limits are documented here: developers.openai.com/codex/pricing

Hadn't hit a Codex rate limit on $20 unless I was running fast mode intensely and as soon as they drop a $100 price tier (the next up), I auto hit the rate limit twice the first day I start using it again sigh

@lordmagus @JustinGorya @rezoundous uhhh sonnet 4 last summer was pretty great actually, lots were done in cursor using sonnet 4. i know we all graduated from HITL workflows and spawn a dozen agents to do mid-level SWE work nowadays but lets not forget what came before lol

@JustinGorya @rezoundous AI was worse then and really could not do what current SOTA does.

So are all $20 AI plans just previews now?

@LLMJunky codex ftw!

@OpenAI the generosity wrt usage resets and promotions from you guys this years has been a godsend. thank you guys for supporting us builders!!!

Our existing $200 Pro tier still remains our highest usage option. And as a thank you to our existing Pro users on the $200 tier, we’re extending our 2x Codex usage promo (until May 31st) and we’ve reset your Codex rate limits (yes, again).

@chalish_b @karpathy @kepano wondering how this tool fair well for handling images and other media in pdfs? always have issues with that, i preferably want the converted pdf-to-markdown transition handle images and tables well and ordered in a similar fashion.

@karpathy @kepano Have you tried Marker? I used it to convert a few textbooks recently and it did a pretty good job. It's a combination of a custom model + LLM to clean things up github.com/datalab-to/mar…

I wrote about Microsoft's Markitdown back in 2024, but it's grown into a big messy project now :/ It would be more valuable if Microsoft provided high-quality official libraries for converting their proprietary formats to Markdown (.docx, .xlsx, .pptx, OneNote, etc). For now Obsidian's Markdown conversion options are: 1. Obsidian Web Clipper for converting URLs 2. Obsidian Importer for converting from apps like Notion, Apple Notes, Google Keep, Microsoft OneNote, Evernote, etc

ehhh... gemini is actually a great model with superb general intelligence, its just the agentic use cases that sucks; really bad at tool calling for instance, particularly in certain harnesses. its not too bad in opencode but even then, open source chinese models like m2.7 or glm 5.1 are better choices in that regard.

there is simply no way that gemini is better than any of these models. it's the most retarded one of them all, by far

@0thernet @steipete @clawcon @zocomputer @openclaw agentic legends!

The ClawFather @steipete made a surprise appearance at @clawcon London 🦞 He's super inspiring (and xtra jacked IRL). My favorite quotes from his Q&A session below ⬇️ PS – my redemption arc is complete, we're on good terms now! @zocomputer ❤️ @openclaw