Moreover, they capture all the smalltalk perfectly, but tend to plausibly mishear the parts that matter most - terminology, abbreviations, names of people. There's no mystery why this happens, of course, but the only transcript I trust is a transcript I personally checked by listening to the audio and fixing mistakes (doable at 1.75x speed, so not that bad). I catch outrageous mistakes sometimes, mistakes that completely change the meaning of what's being said (up to capturing the opposite of what's being said). So, all in all, even though modern speech2text models are really impressive, I am not sure the utility of _completely_ automated transcribers outweighs their dangers today.
The surrender of Wikipedia to indoctrinated gatekeepers is one of the saddest chapters of the Internet history. I hope one day Wkikpedia gets back its own soul.
1) A stakeholder voiced a wish 1 year ago that I forgot about because it was far fetched and not my area. It went to die in a transcript. Today the AI found and resurfaced it in a context where it was actionable and I was in a position to implement it. I did not ask AI to dig it out specifically, like I said, I did not even remember about it, but it just brought it up. Stakeholder ecstatic.
2) I was able to recreate a huge piece of enterprise architecture without access to docs - from 20 meeting transcripts of people just blathering about other topics. From bits and pieces - word here, sentence there - the AI pieced together the big picture for me. Unfeasible manually, especially because the LLM used its training data to infer things that I wouldn't be able to infer myself.
3) I was added to a project as a consulting/observing party, did not pay MUCH attention, but recorded meetings just in case. Suddenly, plans change, and I end up in charge of the project architecture. Nobody bothered to keep any serious records. I throw transcripts, along with emails and chats, into a context window, and I get an excellent self-onboarding doc. I am ready to talk to clients competently next morning.
4) a vendor is failing us, meeting after meeting they defer, delay, gaslight, pretend to forget or misunderstand or not to have heard what they'd been told. I need to escalate, boss asks "give me details to work with". I throw in the transcripts, and in 1 minute I get the timeline of what was going on, with references. Re-tracing this manually would be hopeless.
5) I can talk to a person and ask their opinion virtually, by loading a corpus of meeting transcripts into AI and asking to pretend it's that person. Having 100k tokens of transcripts allows for a pretty high fidelity replica of that colleague!
Dump them in Obsidian with an LLM agent bolted on. This note may never be consciously re-read, but it will become silent part of the context for conversations with the agent in the future. It is _ridiculous_ how useful this approach is.
That pertains to work meetings, though. I would never bring a recorder to a coffee shop.
> maintaining an LLM wiki has been a lot more effort than I thought
Same here. Wikis start out good, but either devolve in a journal-y mess after a while and many updates, or require constant expensive rewrites. (I didn't use the software of the OP.)
You see, this kind of lying and gaslighting is exactly what feeds the distrust in the government and scientific establishment in general public. No number of studies is going to reverse that any time soon.
> Ever wondered what NPR's code of journalistic ethics involves for the newsroom?
I have been thinking a lot through the years about the choice between joirnalistic ethics and journalistic activism in the ranks of organizations like NPR. This is an extremely important topic because today's media are as impactful politically as the "regular" political process.
My point is, such discussion would not make me sleepy, the opposite would happen.
People who like squirrels should get degus. Degus are not propely squirrels, but close cousins, and they are brilliant pets. Their intelligence is outstanding.
Honestly to me it seems that the objectively unprecedented migration levels in the UK of the last decade (caused explicitly by government policy) are better suited to be called "extreme" or "far" than the demands to roll them back
I use Windows 10 with a relatively obscure firewall software with a per-process/per-service whitelist, and try to not be stupid on the Internet. I also do regular backups. This should cover most of the risk model applicable to me. Has worked so far.
I know this is a popular "well actually" to do, but it is not always useful in a conversation. Yes, all cancers are different, but yes, cancer is also one thing: unchecked, harmful division of cells.
Bacteria are also all different, but still they are "one thing", and despite their diversity, antibiotics exist that can deal with many species of them at once. It is reasonable to talk about bacteria and antibacterial medications, it is also reasonable to talk about cancer and cancer treatment. I truly hope cancer will meet its "penicillin" one day (yes I know this is unlikely).
Not the same thing, but I found a way to distribute markdown sources (with images) within the PDF files generated from these sources.
The trick is to generate the PDF normally, then zip this same PDF together with the sources again, with compression level 0, making sure that the PDF is the first file to go in the archive. (Easy to write a script that does this.)
The resulting file, when given the extension PDF, is readable as PDF, and when given the extension ZIP, is extractable as ZIP. So whoever wants the source can rename the file to .zip and extract the source. The instruction to do so can be in the PDF text itself.
Why it works: a) compression level 0 means that the input files are just copied into the stream, so the PDF reader will find the PDF header, decode the rest of the PDF, and ignore the trailing stuff. The trailing stuff contains the markdown sources and the zip directory, making the file a valid archive.
I suspect that tolerances in PDF readers and ZIP decompressors are being slightly abused here, but it works with all PDF readers and ZIP decompressors that I tried so far.