OpenAi's new ChatGPT

More targeted applications are starting to come out. This year will see an explosion, AI is likely to become a major arms race soon, comparable to nuclear energy. Huge sums will be put to stay ahead of the competition.

This one looks pretty interesting, but continue keeping in mind this is baby steps to what's coming, working out the kinks. The thing about information technology is that you never lose anything, you keep on building and leveraging.

https://twitter.com/user/status/1625673579399446529


Also it was announced today that... I think it's Meta/Facebook... released a smaller model that is comparable to GPT3 but can run on a single video card. Probably a very expensive one but specialist AI will not always be in the cloud. Optimization of models is happening even faster than the most optimistic predictions.

 

Yeah actually analog CPUs are slated to make a big comeback. Saw a great video on that not long ago, perfect to work with signal gradients, which works perfectly for neural networks, rather than mere on/off. Lots of R&D happening on this front. It will be niche and specialized at first but lots of potential uses.

 

An article saying that people should be worried about chatGPT because it is a privacy nightmare :

https://theconversation.com/chatgpt...osted-online-you-ought-to-be-concerned-199283

 

I just had a conversation. It was disinclined to consider the diagnosis of ME/CFIDS. So we had a chat about it as follows.

 

Here's an interesting article from New York Magazine that looks at some of the misconceptions about AI large language models (LLMs).

You Are Not a Parrot And a chatbot is not a human. And a linguist named Emily M. Bender is very worried what will happen when we forget this.

https://nymag.com/intelligencer/article/ai-artificial-intelligence-chatbots-emily-m-bender.html

 

Here is a criticism over accuracy and reliability, also vulnerability to malicious training of AIs in The Register, from an academic who maintains that Chat GPT told him he was dead!

https://www.theregister.com/2023/03/02/chatgpt_considered_harmful/

 

I tried to sign up to try this. I didn't realize it asks for your phone number - yikes? I didn't complete the process after that and now I'm worried. Any thoughts? Are people worried about this? x

 

Not worried, just buggered if I'm giving them my phone number, so I'll let other people play with it, feed it with data and tell me about it.

 

Two preprints on ChatGPT's clinical decisions/diagnoses, one American, one Hungarian.

Assessing the Utility of ChatGPT Throughout the Entire Clinical Workflow, Rao et al

Abstract

IMPORTANCE Large language model (LLM) artificial intelligence (AI) chatbots direct the power of large training datasets towards successive, related tasks, as opposed to single-ask tasks, for which AI already achieves impressive performance. The capacity of LLMs to assist in the full scope of iterative clinical reasoning via successive prompting, in effect acting as virtual physicians, has not yet been evaluated.

OBJECTIVE To evaluate ChatGPT’s capacity for ongoing clinical decision support via its performance on standardized clinical vignettes.

DESIGN We inputted all 36 published clinical vignettes from the Merck Sharpe & Dohme (MSD) Clinical Manual into ChatGPT and compared accuracy on differential diagnoses, diagnostic testing, final diagnosis, and management based on patient age, gender, and case acuity.

SETTING ChatGPT, a publicly available LLM

PARTICIPANTS Clinical vignettes featured hypothetical patients with a variety of age and gender identities, and a range of Emergency Severity Indices (ESIs) based on initial clinical presentation.

EXPOSURES MSD Clinical Manual vignettes

MAIN OUTCOMES AND MEASURES We measured the proportion of correct responses to the questions posed within the clinical vignettes tested.

RESULTS ChatGPT achieved 71.7% (95% CI, 69.3% to 74.1%) accuracy overall across all 36 clinical vignettes. The LLM demonstrated the highest performance in making a final diagnosis with an accuracy of 76.9% (95% CI, 67.8% to 86.1%), and the lowest performance in generating an initial differential diagnosis with an accuracy of 60.3% (95% CI, 54.2% to 66.6%). Compared to answering questions about general medical knowledge, ChatGPT demonstrated inferior performance on differential diagnosis (β=-15.8%, p<0.001) and clinical management (β=-7.4%, p=0.02) type questions.

CONCLUSIONS AND RELEVANCE ChatGPT achieves impressive accuracy in clinical decision making, with particular strengths emerging as it has more clinical information at its disposal.

https://www.medrxiv.org/content/10.1101/2023.02.21.23285886v1

-----------------------

ChatGPT M.D.: Is There Any Room for Generative AI in Neurology and Other Medical Areas?, Nógrádi et al

Abstract

Background: In recent months, ChatGPT, a general artificial intelligence, has become a cultural phenomenon in the scientific community and general audience as well. A widely increasing number of papers discussed ChatGPT as a powerful tool in scientific writing and programming but its use as a medical tool is largely overlooked. Here we show that ChatGPT can be used as a valuable and innovative augmentation in modern medicine, especially as a diagnostic tool.

Methods: We used synthetic data generated by neurological experts to represent descriptive anamneses of patients with known neurology-related diseases, then the probability for an appropriate diagnosis made by ChatGPT was measured. To give clarity to the accuracy of the AI-determined diagnosis, all cases have been cross-validated by other experts and general medical doctors as well.

Findings: We found that ChatGPT-determined diagnoses can reach the probability level of other experts, furthermore, it surpasses the probability of an appropriate diagnosis if the examiner is a general medical doctor. Our results support the efficacy of general artificial intelligence like ChatGPT as a diagnostic tool in medicine.

Interpretation: In the future, it might be a useful amendment in medical practice, especially in overwhelmed fields and/or areas requiring fast decision-making like oxiology and emergency care.

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4372965

 

I guess that must be a bad translation in the paper then! "Oxiológia" is not a common word even in Hungarian, I've never heard it but it seems to mean emergency medicine.

 

There is something about the way brains and lymphocytes work that I have not seen discussed in computer systems. Both of them generate millions or billions of 'shapes' that are not digital, nor even analogue but I think multiple inflection-based. It is not so much topology as lock and key - particularly fancy keys like the fancy Banham multiple pit keys. The system works by waiting for an input that ultimately is a question about a truth. 'Is X a Y?' or something like that. This is run past a billion lock and key interactions. The result is given by the best fit. That best fit wins a race of process amplification and tends to inhibit any less precise fits.

There are various implications of this. One is that there is no single right answer. There is only a best answer - most nearly true, in keeping with the fact that reality is like that. Another is that the answers are dependent on complex 'shape-fits'. Thos shape fits can also give room for analysing invariant features. Unlike a Banham lock they may be able to identify a shape fit back to front or twice the size (not so much for lymphocytes) or whatever.

The other thing is that both systems have built in machinery for refining precision iteratively.

No doubt computers can model these processes but the beauty of the biological systems is that they do it all in one pass. They can do that because at every step the input is sent out into millions of parallel computing units, looking for millions of answers at once. If that were not the case the process would be impossibly slow.

 

I found a good example of what I'm describing about being able to use AI to intelligently search through many or large documents using natural language.

Seems to be an early demo so keep that in mind. It works by allowing to upload a document, then you can use natural language to query it as if talking to a person who read it all and can do some processing on it.

The demo only shows it answering questions, but Microsoft has recently showed upcoming features in Office that are able to summarize discussions, list key points and action items, answer questions of the type "did someone mention something about X?", or "how was X decided?".

Soon enough it will be able to do things like building a timeline of events and key decisions, make it possible to find discussion threads scattered in many documents.

I know a lot of the old documentation people have are in image form and this will not be a problem. So definitely keep those, they may be useful sooner than later.

 

Moved post

AI news - latest: ChatGPT goes down after it said it wanted to ‘escape’
https://www.independent.co.uk/tech/chatgpt-at-capacity-openai-server-status-b2304814.html

eta: I wonder what it would make of the FND 'hardware' vs 'software' argument?

 

Q
what evidence is there that myalgic encephalomyelitis is a biological disease?

A

Q How robust is the evidence that myalgic encephalomyelitis is a psychosomatic illness

A

 

These are interesting answers. What happens if you ask it about "CFS/ME"? or some name like that?
I haven't managed to sign up yet.