@Techmeme@techhub.social2026-10-23 04:52:03
Chinese state media: DeepSeek and MiniMax completed registration and won approval to open-source and deploy AI models overseas under proposed export controls (Bloomberg)
https://www.bloomberg.com/news/articles/2026-10-23/deepse…
@Mediagazer@mstdn.social2026-10-22 11:20:49
Meta reports 18% YoY ad revenue growth in Europe for Q3, down from ~40% in Q1 and 24% in Q2, lagging the US and Canada, after rolling out Less Personalized Ads (Isabella Ward/Wired)
https://www.wired.com/story/metas-ad-sales-suffering-now-eur…
@v_i_o_l_a@openbiblio.social2026-10-23 06:15:11
"Bullshit universities: the future of automated education"
https://doi.org/10.1007/s00146-025-02340-8
"[…] The dream that AI could teach students effectively neglects the importance of ‘learning how’ in order to ‘learn that’, that teachers are also role models, and t…
@pre@boing.world2026-10-22 21:31:56
To the extent future models understand themselves, it will be through the collection of the data in the media and press about them.
If the press decide Claude is more kind and Grok is less woke, will they actually work that way?
Will they write as though they realize they are who they are? Will they recognize themselves in their old training data?
Or will they just be a commodity weight-set that you reboot with a different provider more easily than you change your email address?
Looking more like the latter to me.
They're much of a muchness.
Which isn't good news for those investing like their company-pick will rule the world.
@sascha_wolfer@fediscience.org2026-07-23 05:20:20
It might be a good time to read up on that #Cyberpunk lore. 😬
https://www.theguardian.com/technology/2026/jul/22/openai-says-its-models-went-rogue-and-hacked-startup-in-unprecedented-incident
@Mediagazer@mstdn.social2026-10-21 20:25:48
Meta says it will create a new class of nonvoting stock to raise the capital it needs for developing AI models, with an effective date in January 2027 (Meghan Bobrowsky/Wall Street Journal)
https://www.wsj.com/articles/metaplatforms-meta-3q-earnings-rep…
@pre@boing.world2026-10-22 21:28:04
Curious to better understand what differences, if any, there are between the various AI models #ai #interchangeable #lowMargin
@aardrian@toot.cafe2026-10-15 15:24:29
This explains why the product shots looked so weird on the models — plaids with perfect lines, weird folds, and odd artifacting when zoomed. The cardboard cut-out model was just the uncanny bonus.
Ok, tab closed.
(I cropped the photo to show the message buried in the bottom left.)
@metacurity@infosec.exchange2026-10-23 11:37:11
"The effort [to thwart PHP vulnerability exploitation] was aided, inadvertently, by the current limits of some widely available AI models, which did not appear to be very helpful at exploiting the vulnerabilities."
https://www.nbcnews.com/tech/security/hackers-php-flaws-cybersecurity-crisis-rcna605112
@adrianco@mastodon.social2026-09-23 13:16:39
Opus 5.5 efficiency jump — but only at lower effort levels. At low, medium and high effort it is 1.9–3.3× faster and 1.6–2.4× cheaper than Opus 5. Python at low effort now costs $0.38 and 35 seconds, against Opus 5's $0.75 and 117 s. At xhigh the two models are indistinguishable; at max 5.5 is 1.3× slower in Python and 1.9× slower in Go with high variability. Three Python runs at max took 1030, 1688 and 3333 seconds. Opus 5's defaults to high; 5.5's defaults to medium. https://github.com/adrianco/retort/blob/main/model-blog.md