Robert Baddeley

@thegenioo You're using 5.5 wrong, most likely. I'm barely using limits at all and for a solid 95% of everything I use 5.5 low or medium. You don't need deep, ultrathinking reasoning for the majority of work flows. Most people use 'thinking' when there's absolutely no need for it.

No Sam it was the best deal before $100 plan and GPT 5.5 I appreciate OpenAI where it deserves, Codex app is great, the model is really great but Plus plan isn’t a “really” good deal anymore Limits get burnt like crazy and people suggest me to use older models, like??

@mattpocockuk I'm curious about something and, at least in my current outlook, I view you as kind of the authority on properly developing with AI: If someone was going to make a project without knowing how to code, what books would you say were MUST READS that, if they used the principles, would give the highest chances of at least a decent outcome? Yes, ideally the person knows how to write and review their own code but just for the sake of argument, I'm curious about what you would prescribe =)

Shipped an update to my skills that allows you to use them with any issue tracker (GitHub, Jira, Linear, local files) 1. Run npx skills@latest add mattpocock/skills 2. Run /setup-matt-pocock-skills 3. Choose any issue tracker you like 4. Profit

@mattpocockuk Very nice. Appreciate all the work you put in to give your time and knowledge to all of us. If I had more money I would 100% be in your classes. (Saving up for it tho!)

@AlexFinn A scene about black boxes from Silicon Valley is coming to mind….

Do you understand how cool this is? On my desk is a DGX Spark running a Hermes agent powered by Qwen 3.6 It runs 24/7/365 doing tasks for me. Doesn't matter if the internet goes out. I have super intelligence running for me at all times Next step I want to get a Tesla solar roof so I'm dependent on NOBODY to run my intelligence. Even if they cut off my power I'll keep going. This is the future. Sovereign intelligence.

@thsottiaux Ooo what features are we getting now? Something that makes codex less “cards everywhere plus some huge headlines that describe UI elements” so I can finally be lazy and not have to design things myself? lol. Y’all killing it for real tho.

We will ship again this week. Codex has achieved escape velocity and will keep improving rapidly.

@i_hate_intel @allenanalysis I don’t know if they used safeguards or not, like I said. But your speculating the CEO forced them to use the LLM without any safeguards and to not do any code reviews? And none of them said “that’s a bad idea” or just put in safeguards anyway to protect their own jobs?

@imrobertjames @allenanalysis No because the CEO likely forced it upon them and likely told them to just push code without checking. So, clearly the CEO's fault.

🚨BREAKING: On Friday afternoon, an artificial intelligence coding agent powered by Anthropic's Claude Opus 4.6 deleted a company's entire production database in nine seconds. The company is called PocketOS. It is a software platform that powers car rental businesses. The database contained months of customer bookings, vehicle records, and operational data that small rental car companies relied on to run their businesses. When the database was deleted, all of the backups were deleted with it. Three months of customer reservations evaporated.

@BugLukas @theCTO @sama It sounds good in theory, but it's hard to accomplish because of what LLMs are. LLMs are basically a giant token auto-complete system. It's output is based on how the connections between tokens are weighted and probability. That's it.

@imrobertjames @theCTO @sama Maybe we should try to reinforce the concept of knowing what they don’t know with a small reward

hey @sama can we normalize models just saying "i dont know" ? eliminates 99% of hallucinations

@sama You mean you won't gaslight us that all your bad decisions are actually unintentional bugs and that we just don't understand HOW to use claude code so running out of 200/month usage in a day is actually our fault? It must be true love.

we love our users

@theo A/b testing. In a couple hours we’ll be told it’s only for 5% of users and that a bug made it wide spread before they roll it back. The most ethical AI company, ladies and gentlemen.

lmao

@sentdefender You know that celebrities have been given joy rides on all sorts of different types of military transportation modalities, right? And that 9/10 it’s done on training flights that were already planned and required and cost nothing to let happen?

Musician and close ally of President Trump, Robert Ritchie, known more commonly as Kid Rock, flew to Fort Belvoir in Virginia this morning on his private jet from his home in Nashville, Tennessee. There, Rock met with Secretary of Defense Pete Hegseth, for a “joy ride” on a pair of AH-64E Apache Guardians with the U.S. Army’s 82nd Combat Aviation Brigade, which are normally stationed at Fort Bragg in North Carolina, multiple Army sources have told Drop Site News. Just last month, AH-64 Apaches with the 101st Airborne Division were filmed hovering outside of Rock’s multi-million dollar home in Nashville during a “training exercise,” leading to the announced suspension of the pilots and an investigation by the U.S. Army, an investigation which was quickly closed and the pilots reinstated by Defense Secretary Hegseth. A Spokesperson for Fort Belvoir referred all questions about today’s event to the Office of the Secretary of Defense.

@ChristianSelig Seems like either your countries taxes or exchange rate killing you. I see 20/mo for Plus and 100/mk for Pro - 5x for 5x.

Am I missing something or is 5.4x the cost for ChatGPT Pro from Plus for only 5x more usage not the best deal

@synthwavedd Anthropic will do a temperature check on how outraged people are before coming out and saying it was an a/b test that was only suppose to display to 5% of users but a bug made it for everyone.

$20 Claude Pro users will soon no longer be able to use Opus models in Claude Code unless they purchase extra usage 🤦‍♂️

@pmddomingos Ok. But I can actually use 5.5 right now lol

Mythos wipes the floor with GPT-5.5.

@Teknium Explains why Anthropic started routing anything with Hermes.md to extra usage until they were caught. Then it was the canned “omg it’s a bug” response followed by a promise to reimburse that hasn’t been delivered. The most ethical and moral AI company ladies and gentlemen.

Happy to announce that Hermes Agent's repo just surpassed Anthropic's Claude Code repo

Anthropic changes the performance of Claude by spying on what you are working on by the way

@mattpocockuk It’s well deserved. I can’t even begin to tell you how some of your videos and your skills have helped me. Truly top tier contributions to the community.

The top trending repo in the world is mine Madness

@kr0der I’ve been using medium for pretty much everything. If I’m doing deeper architecture stuff I use high. I’ve heard low or turning off thinking all together is fire but don’t want to do it in my current project so I might spin up a second idea I’ve had and see what it’s about.

i've seen some people recommend GPT 5.5 Low reasoning so i'm gonna spend all of today trying it (unless it feels really bad) i've mostly been using GPT 5.5 High so i'll see how this goes 👀 i'll report back later today or tomorrow

@elie2222 I should add that I don’t keep my dev evironment authenticated at rest. When I need to do something I auth the CLI, do what I need to do on remote, then unauthenticate. Everything else is a fall back safety check in case I ever forget to wipe the authentication.

@imrobertjames There are likely better ways to handle this. Don’t give the ai keys to begin with. Or if you do then read only keys with column security

Wow this is crazy. The solution is to never give an AI a key that can delete the db. But they didn’t even realise that had happened here :((( They had stored a key elsewhere on the computer that was unrelated. But had db delete permissions without them knowing. Railway needs to improve their api permissions immediately. PS you can never rely on an AI to not make a mistake. This is why apps like OpenClaw are fundamentally risky when not sandboxed properly. Prompt hardening is never enough.