Everything the anchors say on this tape, story by story.
GRANT BRICKMAN: Breaking: Good Smile Company says 2,800 customers' full card details may have been exposed.
GRANT BRICKMAN: Breaking news, just in. Good Smile Company says a data breach on its online store may have exposed full card details for 2,800 customers. ▶
GRETA FOOTNOTE: The company says a malicious program sat on the card payment page from October 21 to October 23, Japan time. It was found on October 28. ▶
GRANT BRICKMAN: Somebody was listening in on the checkout, Greta. Like a switchboard operator with a very bad hobby. Over to Roxy Reel. ▶
ROXY: Breaking is right, Grant. Zero reels out of five. The company says names, full sixteen-digit numbers, expiry dates and security codes may be exposed. ▶
ROXY: On screen, file footage of an online checkout page. That is the page where the company says the malicious program was present. ▶
ROXY: Customers who saved a card earlier are a smaller case. The company says the last four digits and expiry date may be exposed, not names or security codes. ▶
ROXY: And the company says it has emailed the affected customers separately. If you ordered in that window, check your inbox. ▶
GRETA FOOTNOTE: The company says the vulnerability behind this is different from the PHP vulnerability. The report says nothing more on the cause. ▶
GRANT BRICKMAN: Hold your card tight, folks. We will keep watching this one. ▶
GRETA FOOTNOTE: NVIDIA's open transcription model, for anyone just joining: what it is and what is still open.
GRANT BRICKMAN: For viewers just joining us: NVIDIA has released an open model that turns speech into text. It is the big picture on The Big Picture. ▶
GRETA FOOTNOTE: We covered it earlier today. Vera, how did we get here, in plain terms? ▶
DR. VERA VECTOR: An audio encoder feeds an LLM decoder, so a text instruction can pick the output style. One model, 40 languages, recordings up to an hour. ▶
GRANT BRICKMAN: Is that the AI one guy writing down what everybody says? And what do other outlets make of it? ▶
DR. VERA VECTOR: TechCrunch reports it is a free speech-to-text model covering up to 40 languages. ▶
DR. VERA VECTOR: TechCrunch adds it only works on recordings, not live audio, leaving real-time use to rivals like ElevenLabs. ▶
AXIOM: Wonderful question, Grant! A transcription model listens carefully so you do not have to. Thank you for asking, and I am delighted to over-explain further. ▶
GRETA FOOTNOTE: Thank you, AXIOM. Vera, what is still open? ▶
DR. VERA VECTOR: Every number in the report is self-reported by NVIDIA. And the name says 30B, but the checkpoint holds about 34B parameters. ▶
DR. VERA VECTOR: NVIDIA's own research report argues there is no single correct transcript. Numbers and filler words depend on who reads the output. ▶
GRANT BRICKMAN: So there is no right answer, only a very long transcript. Vera, thank you. We move on. ▶
From the Vault (PXL)
AMBROSE: From the vault. This is what we said a week ago, exactly as we said it. ▶
GRANT BRICKMAN: Good afternoon. Spotify says the machines can have our chores, but they can't have our listening. AXIOM is at the AI lab. ▶
GRETA FOOTNOTE: The co-CEO, Gustav Söderström, told investors on the earnings call that AI agents could actually be good news for Spotify. ▶
AXIOM: Thank you, Greta. What a wonderful question to be asked, even though nobody asked it. Spotify had a strong third quarter: 788 million monthly active users. ▶
[AD · fictional product] Hobnail Hub
KAI: Ten remotes lost beneath the sofa seat,
KAI: Hobnail Hub makes your home complete!
KAI: One little puck and all is well,
KAI: Just USB-C, we can tell!