Tony Onodi
tonyonodi.bsky.social
Tony Onodi
@tonyonodi.bsky.social
87 followers 140 following 48 posts
Developer of numpad.io and meridianapp.co
Posts Media Videos Starter Packs
Reposted by Tony Onodi
I can't understate how good of a job the @inkandswitch.com folks did at this Automerge website. Wow, holy shit automerge.org

And the fact that you can *interact* with the visual demo at the top?

This is the new high bar for a technical website. Holy moly.
Automerge
Automerge is a library for building collaborative, local-first applications.
automerge.org
Interesting. I'm sure the police do it for crimes they actually care about to be honest. I suspect they just wheel this excuse out from time to time when they are under-resourced.
Reposted by Tony Onodi
This is so brilliantly simple, even the police could do it
I built a free tool to try to embarrass British police into doing their job cctv.numpad.io. More info here: onodi.co/bisect/
Reposted by Tony Onodi
This is utterly brilliant:
I built a free tool to try to embarrass British police into doing their job cctv.numpad.io. More info here: onodi.co/bisect/
I built a free tool to try to embarrass British police into doing their job cctv.numpad.io. More info here: onodi.co/bisect/
I read Surely You're Joking Mr Feynman last year and thought "Wow, he really writes exactly as he speaks!" before realising it's one big transcription of recorded interviews
Reposted by Tony Onodi
In general I think it's hard to combat scientific misinformation when some of the best research is locked behind an academic paywall, while lots of nonsense gets published free for everyone to read in predatory journals.
Reposted by Tony Onodi
I thought a MWh battery might be ~shipping container size, or something impractical like that. But with a reasonable-to-low density estimate it turns out to be roughly 3,300 litres, or a cube with 1.5m sides. Not terrible!
I thought a MWh battery might be ~shipping container size, or something impractical like that. But with a reasonable-to-low density estimate it turns out to be roughly 3,300 litres, or a cube with 1.5m sides. Not terrible!
What did non-full-online access look like?
Reposted by Tony Onodi
Since WW2, 19 seats in Westminster have been won with less than 30% of the vote. Ten of those were last year. They'll be a lot more next time.
5 parties on more than 12%, winner gains seat with under 30%. Results like this used to be rare, freak outliers. Now they are routine (there was a second just like this yesterday). Electoral chaos.
Newmarket East (West Suffolk) Council By-Election Result:

➡️ RFM: 29.7% (New)
🌳 CON: 25.0% (+4.1)
🔶 LDM: 17.2% (-3.1)
🌹 LAB: 15.3% (-8.3)
🌍 GRN: 12.8% (New)

No WSI (-19.2) or Ind (-16.1) as previous.

Reform GAIN from Labour.
Changes w/ 2023.
Reposted by Tony Onodi
It’s only just occurred to me that the mantra of TikTok videos is “tell don’t show”
Bohemian Rhapsody Flashmob consisting of 30 musicians and singers on a street in Paris.

I reckon Freddie Mercury would have loved this.

[Sorry ... had to choose the lowest quality and cut into two for BlueSky. Hopefully the audio survived.]

🧵 1/2
Reposted by Tony Onodi
Well fucking do something about it then
Have you noticed the nights drawing in a little earlier?

September is one of the months where we lose sunlight the quickest as we move towards the autumnal equinox
Oh! Yes, sorry. I forgot that wasn’t actually mentioned in this thread.
…so the heuristic GPT-5 relies on is “if I get quizzed about how many letters are in a berry word, the answer is probably 3”
The LLM sees a token for “blue” then “berry” but it can’t see the individual letters. So it presumably has to infer what they are from its training data. I think it gets this wrong because GPT-4 famously thought “strawberry” has two “r”s, and this fact is likely in GPT-5’s training data…
LLMs struggle with this because they can’t “see” individual letters, they get given chunks of letters called tokens. Similarly optical illusions happen because the unconscious parts of our brain process the raw data from our eyes before it gets to the conscious parts and we never see the raw data
LLMs struggle with this because they can’t “see” individual letters, they get given chunks of letters called tokens. Similarly optical illusions happen because the unconscious parts of our brain process the raw data from our eyes before it gets to the conscious parts and we never see the raw data
IMO you’re not allowed to feel smug about weird LLM tokenisation illusions unless your visual system can resolve A and B as the same colour
Reposted by Tony Onodi
John Burn-Murdoch has broken into my apartment and is charting my personal decline across key metrics in red paint over the walls
Reposted by Tony Onodi
There's a realistic chance we get to a position where Wikipedia is blocked while official government accounts are posting on a site that created MechaHitler AI
OSA is gonna make us lose access to fucking wikipedia

write to your MPs.