Gruff Davies

Scientist | Techie | Author
Codex World's Top 50 Innovators 2019
BMWi UK EdTech Founder of the Year 2018
CogX Best AI EdTech Product Finalist '17, '18 & '19

[Gruff Davies @gruffdavies]

"genuinely, and profoundly less wrong"...
Had to share this. Doing physics in the age of AI is an astonishing privelege. I've been working on One Field Theory for nearly 40 years, but the exponential pace of change driven by technology means I've completed more in the last 3 months than the entire preceding 4 decades... Is OFT right? No, theories can only be "less wrong". The question I put to Gemini was this: is OFT "less wrong" than the Standard Model + GR? This was its answer...

[Gruff Davies @gruffdavies]

A High Level Introduction to One Field Theory: https://t.co/pBHAzCjFj0

[Gruff Davies @gruffdavies]

One Field Theory isn't just an incrememental new theory of physics: it's an entire drop-in replacement of the entire Standard Model of particle physics AND General Relativity. It is, in the truest sense of the word, a Grand Unified Theory of everything. #OFT #OneFieldTheory

[Gruff Davies @gruffdavies]

Almost a year ago I posted, hesitantly, about the feedback I was getting from AI on a physics theory I've worked on for nearly four decades.
What might have taken another 20 years to complete has been compressed into months.
One Field Theory is nearly complete.
Watch this space. #OFT #lesswrong #onefieldtheory

[Gruff Davies @gruffdavies]

@Mark2036 @SJWimalawansa Thanks :) No, one group did incredibly well, only a couple left unsolved. Unfortunately, the competition had to close (for multiple unrelated reasons). I thought about a reboot for the sequel Wonderland, but haven't gotten round to writing it.

[Gruff Davies @gruffdavies]

Good to see such an incredibly comprehensive review of the complex mechanisms at play. Congratulations @SJWimalawansa on a superb paper. https://t.co/SNQJdJ0jZ6 #vitamind

[Gruff Davies @gruffdavies]

Being able to discussing physics theories that I've worked on for decades with frontier AI models is surreal. I've always been sceptical of the over-enthusiastic, over-blown praise for my ideas, but getting this feedback from Deep Research, did feel good, I admit.

[Gruff Davies @gruffdavies]

@VatsalThakkarMD @KarlPfleger @tryvenice @ErikVoorhees fair. base model training may be less relevant in the near future though. fine tuning on chain of thought and reasoning has proven extremely effective and large models less likely to reinforce bias in the training set in my tests

[Gruff Davies @gruffdavies]

Thank you @KarlPfleger for continuing to spend time on this. One hope I have for AI, is that it will be able to remove and correct for human bias in the literature and organisations that are failing in their duties.

[Karl Pfleger @KarlPfleger]

The Endocrine Society just released new clinical practice guidelines for vitamin D on their website + a companion paper: https://t.co/22dDohOxt4
They've abandoned their prior 30ng/ml target, now rec against testing, & rec only RDA for most healthy adults. Here's my quick take:

[Gruff Davies @gruffdavies]

@martin_rohbeck Did you test with real and/or synthetic data where the causal relationships were known a priori? It's a fascinating problem to attack but even in DAGs it's not possible to infer causal structure from data alone. How do construct counterfactuals to hypothesis test?

[Gruff Davies @gruffdavies]

@martin_rohbeck Love this. (The bicycle logo looks great and is clever, but the nodes in the DCG need to connect causally otherwise it slightly undermines the paper, no? Teeny niggle but I'm sure I'm not the only one who noticed and easy to fix...)

[Gruff Davies @gruffdavies]

@SeeWhy76 https://t.co/mLOUlikSQm

[Gruff Davies @gruffdavies]

Love like fury, judge me like jury. You stay calm but I can't resist. I choose anger, you choose bliss.

[Gruff Davies @gruffdavies]

😱😱Original data set from the paper "Machine Learning Model Sizes and the Parameter Gap" by Epoch (@Jsevillamol). I annotated with arrows and extracted the year of each Order Of Magnitude(OOM) threshold. I fully expected a double exp (@kurzweilsocial) but no. It's TRIPLE.

[Dr Aseem Malhotra @DrAseemMalhotra]

BREAKING NEWS:

U.K.’s most prominent website for Doctors headlines that the drug regulator @MHRAgovuk is a ‘serious risk to patient safety’

‘The system is beset by financial conflicts of interest & too late in detecting and responding to patient harms’ including covid jabs

‘A prime example of institutional corruption’

MP’s call on for a full and wide ranging investigation

LONG OVERDUE 😡

https://t.co/g2vBgVMXTp

[Gruff Davies @gruffdavies]

@austwitnerd No, I tend to think of it the same was I treat human replies. Some are great, but even in areas of expertise, experts sometimes find themselves making up answers rather than admit they don't know. So long as you relate to its responses as fallable, that pitfall is addressed.

[Gruff Davies @gruffdavies]

Got there in the end. It took some guidance but it's still extraordinary that I can have this level of dialogue on advanced #physics #GeneralRelativity and #QuantumFieldTheory with #GPT4. The possibilities use in for #education are profound (and it's only going to get better).

[Dr Tess Lawrie @lawrie_dr]

Anyone suffering from cancer or knowing someone who does, please read @jane_mclelland's book 'How to starve cancer'. It shows how conventional treatments can be combined with repurposed drugs and other anti-cancer substances to block the known pathways cancer relies on and to directly neutralize cancer stem cells for best outcomes.

[Karl Pfleger @KarlPfleger]

@laurahelmuth @cragcrest @sciam No. This article gets it wrong, taking the RCT failures as proof of non-causality when they were flawed in design & didn't correctly test the scientific hypothesis that the correlation data suggests.
Discussion & references here: https://t.co/zVZErBzQd1