Anthropic, the San Francisco-based AI company known for its Claude chatbot, has filed paperwork for an IPO. The company’s valuation is currently set at 965 billion USD, note that 965 billion USD is not just a small pocket money for firm of 2,500 employees.
1 Like
AI told me the 333 doesn’t have a DAC and the 332 does. Three times and on different days. If it can’t get such basic yet relatively unimportant questions right….
AI is now being used to analyse medical consultations and make notes.
I’m not sure I like the sound of this and how it might impact on the doctor-patient relationship which should generally be confidential - can we be certain it remains so if big business is analysing things?
No doubt financial companies/insurance companies etc will be analysing customer conversations via AI to attempt to detect things they don’t like.
Should things like this not require official consent?
2 Likes
We tried an AI tool to summarise client meetings last year. It was awful.
Tried the same one today; apart from a couple of typos/mis-hearings it was 100% spot on. Spooky. And saves me 30-45 minutes per.
1 Like
The problem being it’s been shown how easy it is for an individual to corrupt what’s summarised.
See Manipulating AI Summarization Features - Schneier on Security .
The summarising of meetings is one of the prime examples of AI proposing a solution to a problem which doesn’t exist and in doing so creating a vulnerability doesn’t exist if done in the traditional manner.
The other being the sifting of application forms. The number of places relying on such an obviously flawed, biased piece of AI does not augur well.
Well in this case it would otherwise be me typing up the summary, so there really isn’t much to corrupt, unless I was corrupt in my own summary. And it saves me half an hour. And, either way, it immediately goes to the client, so…
Us too, and it can also pick up on things missed in a lively discussion, including recording detail like amounts and dates of birth which can sometimes be a problem as you are trying to listen/respond and make notes at the same time.
Copilot has simply revolutionised our business - yes you ALWAYS have to read it through, but it saves a huge amount of time and adds considerable value.
You can then feed your edited meeting note back into Copilot and in seconds it will draft a really well written follow up email, better that I could write.
And even that is just the tip of the iceberg, for example, asking it tough technical questions - in seconds it trawls though our database of reports, tech documents, white papers, the lot and spits out not just the answer but it discusses alternatives, what ifs, things to look out for and so much more, along with references as to where it found the answers, and all this in a few seconds.
And again this is the tip of the iceberg, there are so many possible uses, including comparing a 20-page report against our advice and process standards (200 pages) where Copilot will can spot errors and make suggestions as to how the report could be improved and be made easier to read.
A total game changer and we are finding new ways to use it every day.
I wish I’d run this outpouring through Copilot before posting! 
Shame it didn’t happen 20 years ago.
Tony
There’s an interesting assertion by MS that Co-Pilot respects the boundaries of your business, does not leak your business data onto the net etc. Schneier and many others have shown that this is technically impossible and have given several examples offering proof that it is exactly what it does.
It is ludicrously easy to fall for this clap trap because it “saves time”. MS have been repeatedly asked to explain how it works without taking corporate data onto the internet. Their explanations have been debunked repeatedly.
The desire to sleepwalk into stupidity is phenomenal. As has been noted previously, AI is not after your jobs. Managers are.
1 Like
‘Ambient Listening’ programmes to summarise patient - doctor consultations are here to stay and are transformative. Every doctor I know who uses them says it has transformed their working lives. They tell patients they are using AI and no-one objects. If you use generic ‘free’ AI eg standard Copilot then you will breach GDPR. However, if you are using paid for ‘corporate’ AI then data protected. Most docs I know are using Heidi Ambient Listening which maintains data protection. On the whole it takes 4 iterations of using Heidi to get the record summary that you want - after that it is pretty flawless.
Consider this - GP has 10 mins to see a patient. 7 mins to consult and 3 mins to write notes which they are still doing when next patient comes in. Use Heidi and you have 10 mins for your patient and can look at your next patient as they come in. Bad medicine? I don’t think so
1 Like
This is what I understand:
When you use enterprise-tier Copilot (like Microsoft 365 Copilot), your prompts and organizational data remain securely within your corporate Microsoft 365 tenant boundary.
Maybe someone’s lying to some very smart people in business (on a global scale).
Tony
There’s no “maybe” about it. It’s been demonstrated repeatedly that it’s simply not true.
I think this is not how it plays out at all. I’ll go stronger and say I think there’s strong evidence it’s nonsense and it’s yet another example of the adoption of stupidity.
Multiple GP practices have adopted free and other AI tools and failed to tell patients or give consent opportunities. Implied consent has been wrongly assumed and the BMA agree with suggestions that this egregious stupidity appears to be happening in somewhere between 28 and 50% of cases. Why the lack of precise data? They’ve not been recording use of free tools nor consent to use AI on medical records. It’s rapidly being viewed as a crisis and the ICO has made it clear they’re on the verge of formally stepping in.
Beyond that, my experience both of my own GP and managing staff who work in GP practices, was that in most instances a consultation took as long as it took and ditto the subsequent write up. If I book an appointment with my GP for 9:30am I am fully aware it would be unusual to see them by 9:40am and more likely 9:50 to 10am. The lead GP in the practice is a stickler for 10 minutes but has lost many patients to the other because of that and their many errors and misunderstandings. I’ve never come across a GP who would confine you to 7 minutes and then only use 3 minutes to write up although see SIDEBAR below. Both of mine write up as they go along with you in front of them and continue to write up after you’ve left before inviting in the next patient. I accept YMMV. Some GPs are far more unintelligent than we give them credit for.
I have never come across a GP who invites a patient in whilst they are still writing up the previous case. The potential GDPR breach is obvious.
Heidi itself has massive issues from hallucinating, noting incorrect drug dosages, operating poorly in loud environments and with cheap mikes etc. It has incorrectly classified data and misses subtle verbal clues which can be critical. It doesn’t auto populate your electronic record and text has to be manually copied and pasted. These things mean that checking and correcting often takes longer than if you’d just done it yourself. Assessment of it in practice across Australia showed that all these things meant that in practice it consumed more time than it saved.
SIDEBAR: Certainly came across it with a back surgeon who spent the first 5 minutes of my consult gossiping/bitching about his social life to nurses not involved in my case but stood at the open door whilst he was at his desk also taking a call about another patient and being dismissive about my back scan and the need for surgery. He tried to intertwine questions to me inbetween all of this. I looked at him blankly but he saw no hint to put the phone down or shut the door. I got up, walked out, explained to a nurse that the consultant apparently didn’t wish to see me, went away and lodged a complaint which was upheld in full.
SDEBAR 2: I have just had to email my GP to correct a hospital discharge notice they received from Sheffield which, when I challenged it, turned out to have used AI and was signed by a person who they conceded doesn’t actually exist. The consequence of this improved service was that my GP would have accepted unquestioningly that I was in hospital for 1 night not 4 and was about to issue prescriptions for repeat meds which the hospital said they’d both given me and I’d need for the next 3 months. 4 of those 8 meds were not prescribed at all and certainly weren’t needed by me.
1 Like
I do not doubt your experience or knowledge, but if you are correct this will be the biggest corporate scandal of all time.
I’m not sure it will.
People are only too keen to sign on the dotted line and turn a blind eye to the consequences.
There is a lot to be learned from how MS operate with local authorities on this. You get a cheaper deal the more parts of your authority use Co-Pilot and so LA IT teams are incentivised to encourage use even when it’s hideously inappropriate and there’s zero evidence the claimed safety rails exist.
As I was retiring it was horrific to learn that staff in my service had been told that use of AI tools they knew worked were effectively prohibited because MS could see what you were browsing and see who wasn’t using CP. When I put it to one of our IT team that the guard rails were non-existent and so elected members and officials had been lied to it was suggested that no-one was interested, I couldn’t possibly know what I was talking about, the evidence I supplied was from people with an agenda (well yeah) and if I persisted then disciplinary action might be on the table.
2 Likes