Gareth Dwyer ⠕

You're trying to get your LLMs to stop using em-dashes I'm trying to get mine to stop using purple We are not the same.

Everything going well in Google AI Overview land

(i.e. if I paste an image from my clipboard to Claude Code running in terminal, it should be able to use that image later, e.g. save it to disk, upload it etc etc)

hey @trq212 please can we have a fix for this > ⏺ The image is visible to me in the conversation, but Claude Code doesn't automatically save pasted images as files — I can see it visually but have no path to read or embed it. You need to save it manually.

@badlogicgames fwiw, i'm a codex subscriber and I clicked that link and it said 'off' so I voted 'off' then I realized I'm not logged in on the web. so I logged in and checked again and it was still 'off', but some people might not realize they're not logged in and dirty your data

Quick poll if you are a Codex subscriber. 1. Go to #settings/DataControls" target="_blank" rel="nofollow noopener">chatgpt.com/settings#setti… 2. Is "Improve the model for everyone" on for you?

@wisanishilumani I think they will, even just hooking up Claude to Chrome improves it a lot I think and most people aren't doing that yet

@sixhobbits I sincerely hope they get better. Dark mode stats tiles, no thought to information architecture 🤢

Nothing disgusts me more than a vibe coded UI.

Beginning to think we need current-gen LLMs for tool calling and research, but in a way that they never show any of their 'own' knowledge to the user. Really tired of models arguing with me that recent information/news/models(!) etc don't exist

@zeeg it's hard to balance the takes of the AlwaysHypers (who were also hyped about Web3 and everything else before) and the Dinosaurs (I'll write my assembly by hand it was good enough for software dev in the 60s, why not now)

Why is it everyone with an absurdly futuristic AI take is someone who - as best I can tell - doesn’t work on (and often never has) real software that has real users and real requirements? More so, why do you trust them?

Nobody panic. There’s a South African somewhere figuring out how to make petrol with pineapples.

Seems Snowflake et al are paying Anthropic $$$ for tokens and firing all their writers while Anthropic is paying $$$ for writers. funny old world.

@jdeanwallace there's literally only on ZA bank this could be xD

A software dev friend who works at a ZA bank says he hasn't handwritten a line of code in the last 2 months. They're officially using Claude Code via AWS Bedrock. They even have a leaderboard to track who can use the most tokens. This is a bank 😳

I just forwarded an insurance policy PDF in a language I'm not fluent in to Al, my AI assistant. It found the nasty small print that I suspected was there and now I'll cancel it. Before I probably would have kept it.

contrary to popular belief, I think agents are a net positive in the slop wars. There was plenty of slop before. There's a lot more now. But now my agents can tear through 20k pages and find the treasure. I didn't have that option before.

100% of dev is going to be done in sandboxes in the cloud, controlled by kanban boards. Trust me, I love my local machine and gorgeous mac apps, but all of it is just a terrible form factor for running a team of agents effectively.

Anthropic were the first to notice the flicker. They shipped anyway, thinking nothing of it. By the time we realized it was contagious, it was too late.

@sqs - might be a bit due to random variance but I'm using @AmpCode and @claudeai from Deutsche Bahn spotty wifi. I noticed that Amp seems to not handle the flaky connection as well as Claude. Often getting Timeout with a 'retry' option, then stuck on 'Waiting for response'.

Looking forward to another evening of OpenClaw chats and demos at the AWS Skills Center with the Software Guild luma.com/2wq6dz5r?tk=6V…

Something like this study has been published for every generation of LLMs (from autocomplete to RAG-chat to agents). A few things continue to hold true: - New tools require new training. You don't plop a caveman into a dump truck and expect good results. - That's not to absolve the tool builders from all responsibility—good tools guide the user to good outcomes. - That being said, there's going to be large differences between the top 0.1% most effective (not necessarily the highest token consumer) users and the median user. - New tech creates new problems. The question is whether it is better to embrace new tech and figure out how to mitigate the problems or to avoid altogether.

This is cool 😮 All customer conversations I’ve had when end up on the question of how to manage auth for agents, Kernel is shipping the right stuff here.