iiinova — where your digital self keeps evolving

Richard Gaywood @penllawen.favrd.social · 11/10/2026

Further to the below hilarity: you can see one of the ordered sets in the screenshot. Grok ordered from lego.com, where it's $90. It's $72 on Amazon and Walmart.

140 4

Richard Gaywood @penllawen.favrd.social · 11/10/2026

That set is $90 on Lego.com, which grok used. It's $72 on Walmart and Amazon.

2520 52

Richard Gaywood @penllawen.favrd.social · 26/09/2026

oh no Faine you have activated his trap card

71365 136

Richard Gaywood @penllawen.favrd.social · 23/09/2026

The screenshot below is from one of the memory files that Claude has been keeping for itself on some nonsense I've been vibe coding. This file gets updated at the end of each task, re-read at the start of the next task, then updated again, and so on. How is that different from "learning"?

010 1

Richard Gaywood @penllawen.favrd.social · 23/09/2026

(I would like a fridge that sent me alerts when the temperature was out of range, though, assuming it could be relied on. I had to build that myself.)

A screenshot from home assistant showing temperature readings from four fridges / freezers

Another screenshot showing a 24 hr detailed trace of one particular fridge

010 1

Richard Gaywood @penllawen.favrd.social · 22/09/2026

me setting my camera to 10fps "pro capture" mode: Haha fuck yeah!!! Yes!! me looking at this in lightroom: Well this fucking sucks. What the fuck.

A screenshot of 50 or so thumbnail images in Adobe Lightroom, all of some ducks on the shore of a lake. There's at least 15-20 near-identical shots next to ech other.

030 3

Richard Gaywood @penllawen.favrd.social · 22/09/2026

Here is today's cluesbysam.com, fed into Claude. What do you call this, if not "reasoning through a problem"? It's not guessing at random. It's never seen this puzzle before. How would you describe what is happening here?

100 1

Richard Gaywood @penllawen.favrd.social · 16/09/2026

Here's a more complicated one I made last week of that day's puzzle. This is the complete Claude history.

120 2

Richard Gaywood @penllawen.favrd.social · 16/09/2026

If I were to feed the text of the rules to cluesbysam.com into an LLM accompanied by this screenshot, what do you think happens next? Note that CBS is a daily puzzle game, so the model cannot already know the answers. (This screenshot I have handy is from a few days ago but you could do today's.)

210 2

Richard Gaywood @penllawen.favrd.social · 08/09/2026

so what's the difference between whatever happened here and whatever you think reasoning it seems to do awfully well for a pachinko machine

210 2

Richard Gaywood @penllawen.favrd.social · 08/09/2026

re: LLMs and "reasoning" or lack thereof here's a framing I find interesting if you feed an LLM the "rules" text from cluesbysam.com and then show it the image below, what would you expect it to reply?

A 4x5 puzzle grid from the logic game Clues By Sam. Only one clue is revealed so far: "The only innocent above Wanda is below Betty". The second column contains the following names in order from top to bottom: Betty, Frank, Joyce, Pam, Wanda.

100 1

Richard Gaywood @penllawen.favrd.social · 08/09/2026

So, to be clear, the images do not show reasoning? But if I was solving this puzzle myself and I said these same words aloud and you were listening, you would say that I was reasoning? Or am I not reasoning either?

100 1

Richard Gaywood @penllawen.favrd.social · 08/09/2026

Why do you bring up infinite monkeys? Is it because you still think an LLM can only solve Clues By Sam by guesswork? Please do not misunderstand. That is not my suggestion. I am asking if "picking the correct right move given the grid below" can be done without reasoning or blindly guessing.

100 1

Richard Gaywood @penllawen.favrd.social · 08/09/2026

It's a cracking little game, my favourite of all the daily web puzzle games. I might overthink the weekend puzzles a bit...

010 1

Richard Gaywood @penllawen.favrd.social · 08/09/2026

But Clues By Sam is different every time you play. Look at the grid below. Each new square, new clue, changes the rules by adding or removing the constraints. How do you train a model to play this? What training data exists that lets the model look up the next move here?

200 2

Richard Gaywood @penllawen.favrd.social · 08/09/2026

Random selection would fail very often. There are 38 possible guesses here and only one is correct. Is that what you would expect an LLM to do? If I did this 100 times, how often would you expect an LLM to answer "Betty must be a criminal"?

100 1

Richard Gaywood @penllawen.favrd.social · 08/09/2026

So far we looked at only the first question on today's easy level. Here is a more difficult one. Is figuring out the next move here also "a test for reading" that "does not require reasoning to solve"? Can you solve this just by reading it?

120 2

Richard Gaywood @penllawen.favrd.social · 07/09/2026

Just so I understand If I look at this image and I say "Betty is a criminal" then I have not used reasoning to state that? Using a "typical" definition of "reasoning." If that process isn't reasoning, what is going on in my brain?

220 2

Richard Gaywood @penllawen.favrd.social · 31/08/2026

call it a hunch, but I'm starting to think this 2020-era model ipad pro isn't getting much testing under os 26

A screenshot of an "enter your passcode" prompt on an iPad. It's very, very messed up.

180 8

Richard Gaywood @penllawen.favrd.social · 12/08/2026

can't believe @sistacrumpet.bsky.social didn't put up the pic of our first married tweets tho

me and sistacrumpet, in our wedding outfits, standing back to back, typing on phones and giggling

a tweet from me saying "Figuring out this married thing."

A tweet from sistacrumpet saying "We did!"

170 7

Richard Gaywood @penllawen.favrd.social · 04/08/2026

i read the yegge posts and I was curious what the wyvern players made of all this so I checked the subreddit it's a stream of daily bot posts with a massive wall of text about the changes and zero replies and then occasionally one gets replies and they are all like these

A Reddit screenshot. One user says "Who is drawing these sprite candidates? I hope it doesn't mean what i think it means." (i.e. AI). Another replies "It is what you think it means "

Below that, in a new post, someone else says "Honestly, the addition of Al art usage has killed the game for me. I occasionally/regularly went back for nostalgia sake, fighting through the wall of issues the game has, but I'm out."

Another long post where someone says "So many of these updates sound great, but the endorsement of Al art really drags it down into a state of abysmal bullshit. ... But generative Al is a plague on every creative medium it touches. I haven't ever given a developer a pass for using it, and I don't intend to start now.

"I'll be back when the genAl implementation is gone. Please do better, as genAl is deeply unpopular unless you're in a Silicon Valley bubble."

Yet another user says "Been playing since 2001 .. now the game is truly dead! Fuck off with Al art!"

130 3

Richard Gaywood @penllawen.favrd.social · 28/07/2026

area man who constantly predicts ai will take all the jobs predicts that he is the Most Special Boy and hence the only one who is immune

A Business Insider headline: "Sam Altman says people don't really want an Al to act as CEO: 'I think the world wants to know about the person'"

Underneath that is a picture of his incredibly punchable face.

4115 11

Richard Gaywood @penllawen.favrd.social · 25/07/2026

I have a little claim chowder note just for some of the breathless pundit reactions

110 1

Richard Gaywood @penllawen.favrd.social · 21/07/2026

what in the name of appalling genAI en_gb localisation is going on here

A screenshot of an Amazon listing for a Meross smart gang plug

A screenshot from further down the same listing. There are two FAQs highlighted. One reads "Can I use the Fast Charging PD power strip to charge smaller gadgets without knackering them?" The second says "Looking to charge your gadgets faster while on the move? This smart power strip is your ticket, mate."

1121 12

Richard Gaywood @penllawen.favrd.social · 13/07/2026

yeah that's right found what I think might have been a small misprice on these drives and got them for (still eye-watering) 2023-era prices. they went up ~20% after I ordered

A hard drive upgrade in progress. Three 4 TB WR Red drives are coming out, and three 28 TB Seagate Exos factory refurbished drives are going in.

130 3

Richard Gaywood @penllawen.favrd.social · 12/07/2026

Worth noting that - for whatever reason - Stack Overflow was already in decline long before genAI. I've never seen any definitive hypothesis. It got quite toxic to new users tho. (Highlighted the time from early-mid-2023 onwards when ChatGPT was at 50+M DAU) data.stackexchange.com/stackoverflo...

A graph of Stack Overflow users over time. It peaks in 2014, declines slowly through to 2019/2020, then starts to decline more sharply. The decline speeds up further in early 2023.

2120 12

Richard Gaywood @penllawen.favrd.social · 11/07/2026

i swear to god, this happened just after i read this skeet

010 1

Richard Gaywood @penllawen.favrd.social · 09/07/2026

10G fibre is pretty neat. I sure hope this doesn't awaken anything in me.

A picture of a UniFi network switch with a dangling fibre channel cable connected to it.

010 1

Richard Gaywood @penllawen.favrd.social · 24/06/2026

this made me curious so i made a graph of my london house today consumption of two (inefficient) portable AC units (red and blue, stacked) vs generation from 11 solar panels (green) pretty happy with that and a proper split system would perform much better. i gotta get me one of those

A graph showing my two air con units combined have been approx free to run today as the solar system has generated the same amount of power as they used

182 8

Richard Gaywood @penllawen.favrd.social · 31/05/2026

"inference will ALWAYS become cheaper, what can't be used for inference WILL go to crypto etc, see Jevons paradox" earlier you spoke in absolutes and said Jevons paradox will cause consumption of inference to rise as price falls, no? but now you say "green doesn't always mean go"

100 1

Richard Gaywood @penllawen.favrd.social · 31/05/2026

you said inference providers are in a race to the bottom and platforms will escape that through value add now you jump to content moderation surely content classification/sentiment analysis are simple stateless inference, so why are you bringing that up now? what's the relevance?

100 1

Richard Gaywood @penllawen.favrd.social · 30/05/2026

d'you see that word "if", there? what if it isn't?

010 1

Richard Gaywood @penllawen.favrd.social · 24/05/2026

The week of the release of the puzzle game Blue Prince, I fed a few examples of one part of it (the Parlour Game) into Gemini. It solved them all. It was a wholly new puzzle, invented for the game, couldn't be in the training set. How does a next word guesser guess this?

The rules are:

- there are three boxes - blue, white, black
- you have one key. You can open any one box.
- There will always be at least one box which displays only true statements.
- There will always be at least one box which displays only false statements.
- Only one box has a prize within. The other 2 are always empty.

Then examples of puzzles are

Blue: The gems are in a box with a statement
White: [nothing]
Black: The white box does not have a statement

130 3

Richard Gaywood @penllawen.favrd.social · 01/04/2026

hmmm but it is possible the author of this book is a jackass

A snippet from further down the article

Evidence from beyond DeepMind also removed the space for optimism. In 2019, Google tried to set up its own Advanced Technology External Advisory Council to guide its choices on AI ethics. To achieve political diversity, Google included the president of the conservative Heritage Foundation, Kay Coles James, who had doubts about the advance of rights for trans people. As soon as her appointment was made public, a chorus of social media critics swooped in; the attacks quickly drove other advisory council appointees to withdraw their participation. The public square was dominated by activists who were out to crush opponents, not encourage broad debate. Google's understandable response was to disband the advisory group.

040 4

Richard Gaywood @penllawen.favrd.social · 29/03/2026

absolutely abysmal that i looked at a couple of camera lenses, and now amazon's recommendations algo is spitting out this crap not one of those lenses fits my camera. they're a mix of canon, fuji, sony, and sigma. i shoot on m4/3.

a screenshot of an amazon page section labelled "premium finds for you" with small pictures of six different camera lenses

010 1

Richard Gaywood @penllawen.favrd.social · 28/02/2026

I'll bite.... the week Blue Prince came out, I fed a few examples of one of its puzzles into the then-latest Gemini. (Puzzle rules & examples in screenshots below). It solved them successfully nearly every time. I was quite cross, I wanted to catch it out. Still don't quite know what to make of it.

The puzzle is made up of three boxes that can only be opened using a Wind-up Key. One such key always spawns on the desk, next to the note describing the details:
• There will always be at least one box which displays only true statements.
• There will always be at least one box which displays only false statements.
• Only one box has a prize within. The other 2 are always empty.

240 4

Richard Gaywood @penllawen.favrd.social · 19/02/2026

brb, i'm just gonna make the world's first wallet inspector for the world's first payments platform for ai agents

A screenshot of a tweet from @contraben saying

Introducing Contra Payments.The first payments platform that lets you sell to AI Agents. RT + Comment “Contra” and I’ll send you 100 products AI agents are looking for.

000 0

Richard Gaywood @penllawen.favrd.social · 04/02/2026

openclaw lead dev: "i ship code i never read" same guy being asked to respond to a wide range of vulns found in the openclaw skill registry used by tens of thousands of users: "...got any ideas how? There’s about 1 Million things people want me to do..."

a screenshot of a tweet by @steipete saying "Confession: I ship code I never read"

a tweet from @steipete responding to someone reporting a wide range of malicious code in the OpenClaw skill registry: "Yeah got any ideas how? There’s about 1 Million things people want me to do, I don’t have a magical team that verifies user generated content. Can shut it down or people us their brain when finding skills."

240 4