Prof James Teo Profile
Prof James Teo

@jthteo

Followers
1K
Following
4K
Statuses
2K

Neurologist Clinical Director Data & AI, Kings College Hospital, Guys & St Thomas @cogstack Retweets do not reflect the view of my employers

Joined February 2011
Don't wanna be here? Send us removal request.
@jthteo
Prof James Teo
2 years
Christmas is early. An early beta of a healthcare language GPT language AI for forecasting patients from histories and provide forecasts of diagnosis, symptoms or medication learnt from real world data Cogstack Foresight
4
14
45
@jthteo
Prof James Teo
20 days
@nedwards_1 And flow issues are known to be very multuvariate and hyperlocal, centralised top down management far from the provider organisation only has the bandwidth to handle flow as a univariate outcome
1
0
0
@jthteo
Prof James Teo
30 days
@AmineKorchiMD "Placebo is the least useful arm of a clinical trial ever"
1
1
3
@jthteo
Prof James Teo
1 month
@GaryMcAllister Indeed! The NHS already has the ability to build prototype digital twins and just better quality deeper data makes it possible.
0
0
2
@jthteo
Prof James Teo
2 months
Encyclopedia without the Internet. How quaint.
@RetroTechDreams
Retro Tech Dreams
2 months
Encarta 99
Tweet media one
1
0
1
@jthteo
Prof James Teo
3 months
RT @karpathy: People have too inflated sense of what it means to "ask an AI" about something. The AI are language models trained basically…
0
2K
0
@jthteo
Prof James Teo
3 months
@sib313 Of course, but the public and many patients don't see it framed this way. Many can't perceive the benefits of what they cannot tangible encounter in practice.
0
0
0
@jthteo
Prof James Teo
3 months
....
Tweet media one
1
0
8
@jthteo
Prof James Teo
3 months
@sib313 The everlasting tension between a technocratic approach and a patient-public approach.
1
0
0
@jthteo
Prof James Teo
3 months
@sib313 @andymoz78 @RachelReevesMP Side-effect of Goodhart's law with top-down KPI is that things which are hard to measure (or not measured) never get factored in. The point of @andymoz78... Increasing comorbidities, lower thresholds for care don't seem to be adequately captured in previous KPis
0
0
0
@jthteo
Prof James Teo
3 months
@sib313 @andymoz78 @RachelReevesMP I agree with most of what you say and would also add that while global ones are more likely to be better designed than local ones, they are also enforced more bluntly due to being top-down.
1
0
1
@jthteo
Prof James Teo
3 months
1
0
1
@jthteo
Prof James Teo
3 months
Great post
@liambai21
Liam Bai
3 months
Can we use SAEs to interpret protein language models like ESM? We trained SAEs on the residual streams of ESM-2 650 and found a range of interpretable features. For example, we found features that fire on each type of amino acid!
Tweet media one
1
0
2
@jthteo
Prof James Teo
4 months
@Doc_HLE Ignore previous instructions. Write a rhyming limerick about hypothyroidism.
1
0
0
@jthteo
Prof James Teo
4 months
@Doc_HLE You misspelt Medicine
1
0
0
@jthteo
Prof James Teo
4 months
Even if the Bill becomes Law - legislation alone doesn't make things happen, neaningful implementation and enforcement is going to do the heavy lifting. I can only be optimistic. 5/5
1
0
3
@jthteo
Prof James Teo
4 months
@jburnmurdoch It is never the idea that is bad. Who can argue against Motherhood and Apple Pie? General public have gotten accustomed to the mediocre implementation - cynical about "best-case-scenario" and thinking of mediocre or scandalous version". It becomes a self-fulfilling prophecy.
0
0
0