Beep boop - this is a robot. A new show has been posted to TWiT…
What are your thoughts about today’s show? We’d love to hear from you!
Beep boop - this is a robot. A new show has been posted to TWiT…
What are your thoughts about today’s show? We’d love to hear from you!
I think @Leo may have missed a trick wrt the marmots–it being Feb 2nd and all.
Is it pronounced MAR-MOT - or, like the french pronounciation - MAR-MO?
The one problem I have with AI coding, without humans checking the code, is that one of the biggest problems we’ve had, as programmers, is that nobody can really define what the code needs to do.
I worked from the 80s through to the 2010s as a programmer and the quality of specifications went seriously down hill after about 2002. It went from clear and concise, but incomplete, to almost back of a fag-packet (back of a napkin, for my American friends).
But even in the days when specifications were really detailed, they still missed a lot of things that good programmers just knew. They’d ask for a piece of code that asked the user for information and write it to the database. It didn’t tell the programmer to escape the input, check the size of the input for buffer overflows etc. Good programmers know to automatically put in those checks, a bad programmer blindly follows the spec.
Given the quality of the specs has gone downhill, in my experience, I don’t see AI coders being able to think up all of those hundreds of missing pieces for each instruction they do have. And, at least until AIs stop hallucinating, you won’t be able to remove the experienced coder from the chain, at least to confirm the code is correct, for anything that will be used commercially.
Yes. I was being funny. I still forget sometimes how humor doesn’t translate over the internet.
I had glanced at few minutes of the live stream of this episode on Sunday; I made a bookmark to go back. Leo had commented on spending ~$200 per month on AIs, so I scanned the transcript for that comment. No luck. I found where the transcript noted Leo’s words @36:19: “I am an AI accelerationist.” The transcribe missed the next comment completely: “Literally, you know, probably 100 to 200 bucks a month.” That sentence was completely absent from the transcript!
What’s jarring is that the AI transcriptionist gets difficult things right – like accelerationist – and then totally botches it on simple stuff. Most have some gauge for what should be easy or hard, but I am constantly thrown off by the failures of the AIs.
Christina then mentioned @38:21 that AI startups that intend to provide take body cam footage and “transcribe things”. Again the AI totally skipped over the words transcribe things. It’s quite amusing that those two words got dropped out of the transcript. Unless that AI has some bulletproof way to cross-check every freakin’ oscillation recorded in the BodyCam audio, I cannot imagine AI-generated BodyCam transcripts being legally admissible. Some human would always have to check the recording. I’m betting most professional transcriptionists use an AI to start their transcript, but there’s no way that an AI can substitute for an attentive human. I don’t think that Elaine’s gig with Security Now! is in any jeopardy.
Marmots are awesome. There are marmots and pikas on the Long’s Peak trail in Boulder County, Colorado. Marmots are at the summit of Long’s Peak (14,259 feet). It rather astonishing to see them in the desolate boulders and smaller rocks up there. Maybe Henry did this marmot-pilgrimage when he was at CU.
I found a fun online conversation: What do Marmots eat at the top of Long’s Peak? The TL;DR answer: Pop Tarts. Trigger warning: there are some squeamish details in that conversation.