AI Summary

[formoverfunction]

I've been using https://www.investormeetcompany.com/ almost since it started, mainly for the presenations. I'd rather pick up my RNS, which is available on the site, directly from the LSE News Explorer:

https://www.londonstockexchange.com/new ... etypes=1,2

But, while I was waiting for a presentation to load I noticed they were showing AI technology on the splash page, so I thought I'd have a quick look.

I was interested to see that it's actually a summary of RNS via AI, under the RNS feed tab:

https://www.investormeetcompany.com/news-items

An example, this morning's Interim Report for IPO the AI churns out:

"AI Summary

IP Group plc reported good progress in its half-yearly results for the period ending June 30, 2024, with a focus on cash proceeds and portfolio development. Key highlights include:

Cash Proceeds: A total of £3.4 million from disposals, with an additional £41.2 million expected in the second half (including proceeds from the sale of Garrison Technology for £30 million).

Investment Activity: £49.1 million invested across 23 companies in high-growth sectors, with £380 million raised in the portfolio during the period.

Financial Position: NAV per share decreased to 104.7p (down 9% from the previous year), with an overall loss of £109.9 million due to a significant decline in the market value of listed holding Oxford Nanopore.

Balance Sheet: Maintained a strong cash position of £161.3 million at period-end, later increasing to £183.7 million.

Share Buyback Program: Continued share buyback amidst a price discount to NAV, completing £8.1 million of a £20 million plan.

Portfolio Developments: Positive clinical data reported from life sciences companies, resilience in deeptech with strong growth in cybersecurity firm Featurespace, and significant funding raised in cleantech.

The outlook remains positive reflecting improving market conditions and confidence in future exits."

I thought it might be interest when it's a busy reporting morning, keeping notes or first stage research.

Overall, I'm very impressed with the site and the way it's developing. At first I was concerned about privacy and it generating junk mail, so I gave it's an aliased email, but nothing. I'm impressed enough to have shelved my own plans to try and use AI to make summary reports.

[Gilgongo]

I thought it might be interest when it's a busy reporting morning, keeping notes or first stage research.

I get the impression text summaries are one of the few areas where AI is having a proper impact on productivity. I've used them for news articles too. You do have to keep your wits about you a bit as sometimes as you'll read something that sounds fine at first glance but doesn't quite feel right. And when you go to the source you see it's mixed something up. But that's rare and possibly getting rarer now.

As an aside, I used to be a language translator before Google Translate came to that field. I've not done professional translation since then but understand these days the pros will do a first-pass machine translation to use as a "rough", then refer to the original to sort out issues of style, ambiguity or the inevitable plain mistake. Maybe back translate it for a quick check (because the client will!), then email it off. The downside is that I was charging £200 per 1,000 words back in 1997 whereas now it's about £20!

[formoverfunction]

I've found the summaries quite useful, I still read the full text, then the summary. I keep a local version as they are on average well under 100k, so it's useful for notes, watch lists etc.

I haven't yet tried it, but the idea of having AI summeries video presentations appeals.

[GoSeigen]

There's not the remotest chance I'd use AI summaries for my investing purposes given the current state of the art except in some rare cases when I can check the output thoroughly. Complete waste of time. I'd never know when the information was reliable or completely fabricated and would have to refer to the original to verify every fact. How does that help?

I recently asked an AI to help tidy up some OCR text. It went beyond the brief and started removing bits and adding in others. Without thoroughly checking you just don't know what the output will be -- and I expect all liability for consequential loss is excluded in the contract.


GS

[OhNoNotimAgain]

For the first time ever I agree with GS

[Whatsitworth]

Iv started noticing a lot of mistakes, particularly misplaced or misspelled words in articles lately. Is that them using A.I to write them? I read one today about Burberry and twice they used mulberry instead of Burberry.

[bofh]

There's not the remotest chance I'd use AI summaries for my investing purposes given the current state of the art except in some rare cases when I can check the output thoroughly. Complete waste of time. I'd never know when the information was reliable or completely fabricated and would have to refer to the original to verify every fact. How does that help?

I'm a very heavy AI user, and I'm getting a lot more done within domains that align well with my expertise. Since I can quickly vet the responses (and as was noted, you have to vet the AI's answers), AI is proving to be a material productivity booster. Moreover, it increases your appetite to take on tasks you previously might have deferred due to the opportunity cost. Again, I'm referring to tasks where I have competency and can vet results.

For investment purposes, I'd want two things as a starting point: trustworthy citations and LLM tooling that can run financial computations (i.e., a non-deterministic LLM calls out to a deterministic tool to calculate results). The latter is already offered in ChatGPT, but deterministic quoting/citations are a work in progress.

Beyond that, it would be nice to state risk appetite and preferred investment principles or rules and have the LLM observe those (this can "mostly" be achieved with prompting today).

There's a significant demand for trustworthy citations in legal and healthcare. Reliable techniques to achieve these are already emerging. The killer app will have a 100% mistake-free guarantee around citations and computations for regulated organisations. I worded that carefully since the LLM can still make things up, misdirect, or draw wrong conclusions even with perfect citations and calculations.

It's good to remember that LLMs are dream machines, and someone smarter than me already noted that "hallucinations" are just the dreams you didn't ask for.