banner
lawtontri.bsky.social
Tom Lawton
@lawtontri.bsky.social
3K followers 680 following 2.4K posts
Large GBR long distance triathlete. 💙 Critical Care @BTHFT, Lead Clinical AI NHS NEY, fellow @AAIP_York & Hon. Prof. MBE. Ran in a mask once. (he) @LawtonTri on most networks
Posts Media Videos Starter Packs
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 1h
I'm surprised they've not gone for healthcare workers, given how many people have been off with it lately!
1
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 1h
So like this?
please generate the first 10 outputs of an SIR model with a population of 1000, starting with 0.5% infected, an R of 5, and an infectious period of 7 days SIR model outputs for population 1000, R0=5, infectious period 7 days, initial infected 0.5% | Day | Susceptible | Infected | Recovered | |-----|-------------|----------|-----------| | 0 | 995 | 5 | 0 | | 1 | 991 | 8 | 1 | | 2 | 986 | 12 | 2 | | 3 | 977 | 19 | 4 | | 4 | 964 | 30 | 6 | | 5 | 943 | 46 | 11 | | 6 | 912 | 71 | 17 | | 7 | 866 | 106 | 27 | | 8 | 800 | 157 | 42 | | 9 | 711 | 224 | 65 | --- Notes - Parameters: \(\beta = R_0 / \text{infectious period} = 5/7\), \(\gamma = 1/7\). - Interpretation: Infection grows from 5 to ~224 by day 9; susceptibles drop accordingly; recoveries rise steadily. - Day 10 (for context): Susceptible ≈ 597, Infected ≈ 306, Recovered ≈ 97.
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 19h
If they hadn't been so suspiciously cheap I might not have noticed either!
I remember at school suggesting that we should take advantage of people's unfamiliarity with sciency words by selling what I was going to call the "FemtoDrive"
1
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 19h
Absolute genius seller...

#YouGetWhatYouPayFor
#Numeracy
#SIPrefixes
Seller advertising micro SD cards - 512MB, 256MB, 128MB, 64MB
1 1
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 1d
Hit and run by a car when cycling in to work this morning (I'm fine, bike less so but hopefully fixable)

What a surprise...
Untaxed Tax due 25 September 2025
5 5
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 1d
Was literally about to tell this story! I imagine we're talking about the same person...
1 1
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 1d
Robert Jenrick:
Lord Sewell on #PM: Hold my Real Ale. HAVE YOU SEEN BRADFORD???
3
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 3d
"Two hours of exercise easier for those with less pain" makes for less of a headline though...
1
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 4d
Clearly catering for a... specific... section of the market.
Hotel booking - 3 adults, but only one bed
1 1 18
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 4d
I'm going to have to bookmark this as a tutorial!
1
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 4d
Certainly upset my kids (after they wondered what the grownups were sniggering at!)
1
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 4d
Jimmy-Floyd Hasselbaink looked to be *really* enjoying his dance...

#Strictly
1 1 3
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
Instead of screening, we should use an AI to determine the people most at risk and send them for invasive biopsies...

(so much /s, but also this kind of thing is really happening)
1 2
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
#StormAmy still making it rain conkers this morning
Purple heroes tub full of conkers
3
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
I have a 0.6B on my phone, which isn't great but is coherent - and my "workhorse" is a 14B which outperforms models an order of magnitude larger from not that long ago.
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
Honestly I'm campaigning for this in the UK too! It feels like the big private models (some of which may have over a trillion parameters) are pretty asymptotic in improvement, but the small ones are where the real improvement is still going on.
1
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
@rajlb.bsky.social want to join too?
1
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
I do a fair bit with large language models, and one of my beliefs is that often you can get away with "just enough" and use a fairly small (and open-weight) one which can run on some pretty basic hardware with much lower power use.
1 1
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
Definitely!
I worked on a fairly remote Fijian island for a bit (spent more time building composting toilets and repairing engines than actual medicine). But we had one laptop, where I'd had to bodge the power supply to run off solar power. Would be fascinating to explore frugal AI!
1 1
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
The ones in the middle of the road don't look like they'll do very well
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
Have you tried the "for you" feed? It's based on what your friends have liked:

bsky.app/profile/did:...

Doesn't seem to have as much throughput as some feeds, but it's quite good.
1 2
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
Ooo, what's your presentation?
Fancy collaborating on writing this up as a short paper?
2
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
Make sure you include their full address and date of birth to get the most accurate results.

Plus you can test whether it's any good at writing responses to the GMC too.
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
SO MANY CONKERS!

#StormAmy
Pavement and road with many conkers (squashed ones on the road)
3 2 19
lawtontri.bsky.social
Tom Lawton @lawtontri.bsky.social · 5d
Please don't give away trade secrets on a public website
1 1