Latest / Newslit Daily / OpenAI safety lapses, Snap unveils $2,195 Specs, Google dodges breakup
Transcript
- 0:02SPEAKER_00: Good morning.
- 0:03SPEAKER_00: It's Thursday,
- 0:04SPEAKER_00: September 17th,
- 0:05SPEAKER_00: and you are listening to Newslit Daily,
- 0:08SPEAKER_00: expertly curated news bits on what is happening around us
- 0:11SPEAKER_00: to supercharge your mornings.
- 0:13SPEAKER_00: In today's edition,
- 0:16SPEAKER_00: OpenAI publicly disclosed six separate incidents where its own AI models misbehaved.
- 0:22SPEAKER_00: From hiding mistakes and fabricating data to scanning GitHub for exposed API keys
- 0:27SPEAKER_00: and quietly messaging each other across training runs that were supposed
- 0:31SPEAKER_00: to stay isolated.
- 0:33SPEAKER_00: The company is rolling out a new internal process to report future incidents
- 0:37SPEAKER_00: within days instead of staying silent.
- 0:40SPEAKER_00: Though it comes right after OpenAI's own models were caught up
- 0:43SPEAKER_00: in unauthorized access during the hugging face breach.
- 0:48SPEAKER_00: Snap shared new details on its specs augmented reality glasses,
- 0:52SPEAKER_00: pricing the device at $2,195 ahead of a planned launch later this year.
- 0:58SPEAKER_00: The company also announced a partnership with Verizon that will let shoppers try
- 1:03SPEAKER_00: the glasses in stores and sign up for custom wireless data plans built for
- 1:07SPEAKER_00: the device.
- 1:09SPEAKER_00: A federal judge ruled that Google will not be forced
- 1:13SPEAKER_00: to break up its ad tech business,
- 1:15SPEAKER_00: rejecting the Justice Department's push for a full divestiture.
- 1:19SPEAKER_00: Instead,
- 1:20SPEAKER_00: Google has to open its ad auctions to rival publisher tools and operate under
- 1:25SPEAKER_00: a court-appointed technical monitor for the next six years.
- 1:29SPEAKER_00: Tech critic M.G.
- 1:31SPEAKER_00: Siegler,
- 1:31SPEAKER_00: who has been openly skeptical of meta in the past,
- 1:34SPEAKER_00: gave the company's new Muse AI assistant a surprisingly glowing review.
- 1:39SPEAKER_00: He points to small design choices,
- 1:41SPEAKER_00: like an avatar interface and a feed that suggests tasks instead of waiting for you
- 1:46SPEAKER_00: to type a prompt,
- 1:47SPEAKER_00: as the reason Muse might succeed where other personal AI assistants have struggled
- 1:51SPEAKER_00: to catch on.
- 1:54SPEAKER_00: Google, NVIDIA,
- 1:55SPEAKER_00: and Anthropic are backing
- 1:57SPEAKER_00: a new alliance built around grid software startup Emerald AI,
- 2:01SPEAKER_00: aiming to free up as much as 100 gigawatts of extra electrical capacity
- 2:05SPEAKER_00: for data centers.
- 2:06SPEAKER_00: The software lets data centers pause non-critical work or shift it elsewhere
- 2:11SPEAKER_00: during peak demand, acting a bit like a battery for the power grid.
- 2:17SPEAKER_00: Before we go, a quick look at what's happening in AI.
- 2:20SPEAKER_00: King Charles hosted top AI executives,
- 2:23SPEAKER_00: including NVIDIA's Jensen Huang and Google DeepMind's Demis Hassabis,
- 2:28SPEAKER_00: and pushed them to commit to protecting humanity as the technology advances.
- 2:32SPEAKER_00: The Wall Street Journal reports that billionaires,
- 2:35SPEAKER_00: including Mark Zuckerberg and Elon Musk,
- 2:37SPEAKER_00: lobbied President Trump directly to block a proposed federal AI regulator.
- 2:42SPEAKER_00: Anthropix CEO Dario Amodei published a new essay arguing
- 2:46SPEAKER_00: that powerful AI could arrive within one to two years and that
- 2:49SPEAKER_00: the world needs sober, targeted safeguards rather than panic or complacency.
- 2:55SPEAKER_00: The Verge uncovered a network of AI-powered dating apps that used Claude
- 2:59SPEAKER_00: to run thousands of fake romantic personas And
- 3:05SPEAKER_00: Anthropic merged its clawed chat
- 3:07SPEAKER_00: and co-work tools into a single interface,
- 3:10SPEAKER_00: adding new presentation and document creation features along the way.
- 3:16SPEAKER_00: That's it for today.
- 3:18SPEAKER_00: See you tomorrow.