Matthew Hooson (@MrHoosk) profile picture

Odd jobs for tokens

Manchester
Joined January 2018
Photo by Matthew Hooson (@MrHoosk)Photo by Matthew Hooson (@MrHoosk)Photo by Matthew Hooson (@MrHoosk)Photo by Matthew Hooson (@MrHoosk)Photo by Matthew Hooson (@MrHoosk)Photo by Matthew Hooson (@MrHoosk)Photo by Matthew Hooson (@MrHoosk)Photo by Matthew Hooson (@MrHoosk)Photo by Matthew Hooson (@MrHoosk)
"Agentic": 67 posts a day in January 2023. 8,155 now.
1
AGI is now declared on X about 44 times a day. The best AI model scores 64.6.
2
"As an AI language model" peaked at 105 posts a day on X in April 2023. It's down to 5. The bots learned. (or the posters did)
3
In October 2022, about 27 posts a day on X asked "is this AI?" Now it's 1,352.
4
Continuing my 'delve' into how LLMs have impacted twitter/X since they bust onto the scene, the work delve looks to finally be returning to normal use after a surge post Chat GPT launch. The same thing was seen in Scientific publishing! A word the models use much less now.
11
To prove my point on the out of control use of the word 'insane' here is its usage over time mapped when used alongside AI terms
1
1
36
The rise of AI has a lot to answer for, not least of which is the explosion in everyone saying 'insane' all the damn time. Let's branch out everyone, this is insane
1
20
On 24 July, Claude Opus 5 was the top-scoring AI model in the world. 112 days later it ranks 13th. Twelve newer models have passed it. The latest, Gemini 4 Hydrogen, came out yesterday.
1
1
180
Since ChatGPT launched, 32 different models have held the top spot in AI. 29 of them were made by OpenAI or Anthropic. Google has done it twice. SpaceXAI once. Meta, DeepSeek and every other lab: zero. (Based on intelligence index, not real world use)
1
1
95
How far behind is each AI lab? Take a lab's best model and ask: when was that last good enough to be No. 1? Google: 2 days ago OpenAI: 50 days ago Meta: 69 days ago, and its model only launched last month
1
19
AI: A Bubble with Benefits Better models, bigger compute bills and valuations that take some explaining. Dara and Matthew ask how much AI progress is useful, and how much investors are paying for. Watch youtu.be/irJLTdh3gPc Listen open.spotify.com/episode/5vF…
1
1
18
Some funkiness when @claudeai code context window fills up. Rather than compacting automatically, it just stops and prompts me to compact. Not sure if a Fable 5.2 thing, but only noticed it since using Fable today
21
This is something that keeps me up at night at times. Trying to review code the old way in the face of a crazy volume of changes. I have definitely come to lean on Claude to review its own work more and more and may well try out the flow from the bigwigs themselves claude.dev/blog/solving-code…
1
17
I reckon this is the pre cursor to wider dots roll out... I am not sure I actually want a dot, but I will be damned if the choice not use it my own!
Fellow resetters, heads up that we're working on expanding the family of frontier models that power dots, including GPT-6 Astra and GPT-6.1 Terra and some others in the future. Terra brings speed and a nice personality upgrade, and will help us bring dots to more people.
18
I'm all for protecting hard working game devs etc, but there is a fine line that needs to be walked here. For example I am not all for the monopolistic practices adobe has wielded for way to long, and am happy to see their chips peed on a little
An update on building games with Claude Code: Last week some of you ran into Claude Code pushing back on game projects it should have just built. We'd added a short rule to the Opus 5.5 prompt to steer Claude away from recreating specific existing games, but it was also kicking in when you referenced one just to describe the mechanics you wanted. We've updated that rule in 2.1.317. You can name a game to describe what you want and Claude should now build an original game in that genre. If you ask for a copy of the original, it'll offer to build something similar instead. We want Claude to be as helpful as possible when you're building something original, but also to protect your work if someone else tries to copy it. We'll keep improving this, so let us know if Claude turns down work that's yours. Thanks to everyone who shared feedback!
1
19
I find it hard to envisage where the models can go from Opus 5.5. As a model to chew through coding tasks over long sessions, it feels near perfect. (Cut to me being embarrassed by this post when Opus 7 is rewiring my DNA to make me more tolerant to dairy)
17
I don't think I see enough people talk about Claude Mods. When you are limited in terms of an interface or control plane (maybe by the company you work for) being able to extend and customize is a game changer code.claude.com/docs/en/plug…
18
I really need to understand what the hell these elements mean in Google's new models. I argon the middle model of the Gemini 4 model family? Is carbon the top model? Or is carbon a replacement for Argon and if so, what the hell is the 4 (other than a really good HTTPS cert key)
30
Bots, Bugs and Business Models Episode 3: custom tools, small-business creativity and the security holes AI is finding. Plus dangerously high fizz. Watch: youtu.be/AKXzS6CfLso Listen: open.spotify.com/episode/0I2… #AI
1
17
Just one more AI experiment. In Episode 2, we talk agents, Paperclip, model routing and the temptation to keep tinkering. Plus the odd tangent and a fizz update. Watch: youtu.be/aiq6JmURXrY Listen: open.spotify.com/episode/1ep… #AIAgents
1
1
1
467