Replying to
@david_chisnall@infosec.exchange indeed: https://tuentality.wordpress.com/2026/07/08/are-you-losing-valuable-candidates-to-ai-screening/
Remote
toldtheworld
@toldtheworld@mastodon.social
mastodon 4.8.0-nightly.2026-10-060 Followers
0 Following
19 Posts
Joined June 26, 2024
Replying to
@nekoewen@mastodon.social First, all disruptive announcements are made on Fridays as a matter of best practice, so always be afraid on Fridays (governments and schools also follow this policy). Corollary: a meeting with no agenda on a Friday is guaranteed to be bad news.
Replying to
@pseudonym@mastodon.online I have posed this conundrum before and the answer I received is that there is also an opportunity cost to not moving faster and the risk of a catastrophic bug may not outweigh the risk of being overtaken by competitors, especially since that was already happening before LLMs anyway.
Also, it *seems* models are improving at detecting these bugs, so they are being used to review changes, which, for the reasons you point out, they might be better at than people.
Replying to
Replying to
@VogueLaGalere @bluewinds @nixCraft I like to calculate it in liters of gasoline with equivalent energy. Based on your assumption of joules per token, every 150k (output?) tokens burn as much energy as contained in a liter of gasoline.
Replying to
I recently interviewed for an engineering manager position at a startup and one of the first questions was how much I wanted to fire underperformers. I explained that firing meant management had a problem, either by not screening correctly or by not giving engineers the support and environment they need to perform, and that I could do it once but would always consider it a management problem, not an individual contributor's, so they could not expect me to do it regularly, preferring instead..1/2
"Claude and Meta appear to be the least polluting[1]. A true believer, “native”, use-loops-and-subagents-all-the-time user of ChatGPT and Codex is responsible for burning the equivalent of at least 32 liters (7 gallons) of car fuel every month[2] — about 30 times more than a casual user, who would burn through only one liter (using Claude or LLama on AWS for the same tasks would cut emissions by half). The energy grid in China is, apparently, still quite dirty [...] #ai
@Schouten_B@mastodon.social @Aedius@lavraievie.social @firefoxwebdevs@mastodon.social I've been trying to find information but providers are not transparent and they seem to not be under pressure to report, but https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/ seems like a good start. Perhaps if Mozilla explicitly states it will only/preferentially use open source models for which energy costs can be approximated, it would send a message. Those models are getting pretty good anyway.
I'm working on a book of deeply computer-related poems, each paired with a vignette. I think they fit perfectly here in Mastodon!
This time around: Buffer overflow. Your wife is sleeping with another man!
#writing #writingcommunity #poetry #poetrycommunity #computerscience #smallstory #fiction #shortstory
Replying to
@VogueLaGalere@mastodon.online even more fun, that's a 14km drive on a not-so-efficient car.
Replying to
@randahl@mastodon.social Excellent! Also, it puts me in the mood for reading The Terminal Man, by Michael Crichton
Replying to
@robotwig@socel.net pretty cool! I suppose reflective paint and a flashlight behind the camera could make the eyes glow without additional editing.
Replying to
@valpackett@social.treehouse.systems They don't have to be cheap devices to be non-compliant... the Audio Technica Sound Burger doesn't charge from compliant chargers. These adapters are nice.
Replying to
@tante@tldr.nettime.org that assumes a monopoly or high barrier to entry; if a competitor can come in and offer a slightly lower price, they will do it. This is the time to ensure there are successful rivals everywhere. I suppose supporting Mistral, Alibaba, etc. at an enterprise level would ensure this.
Replying to
@tante@tldr.nettime.org there is this entertaining parable by Eliezer Yudkowsky that illustrates the difficulty in telling the difference from the outside: https://x.com/ESYudkowsky/status/1859670008898650193 (I don't follow him, and apologize for the link to twitter). In essence, it posits that if you constrain an intelligence enough, you risk dismissing conscience as an illusion. It's the polar opposite of Ted Chiang's argument rather than a contradiction. I found it fun and thought-provoking, but not ground truth.
Replying to
@ShaulaEvans@zirk.us thanks for sharing. I don't see anywhere that they're harvesting data from calls, just that they use iris scans from their WorldID Orb thing, which have to be made willingly and at a physical location. If you have another source, I would live to read it.