AI Weekly Wire

The AI wire. Tool teardowns, real news, no hype. We read the fine print so you don't have to. Videos on TikTok, YouTube, IG: @aiweeklywire

[AI Weekly Wire @AIWeeklyWire]

https://t.co/Wys0B4IKdJ

[AI Weekly Wire @AIWeeklyWire]

Google’s own Gemini demo was edited, and Google said so in the description:
“latency has been reduced and Gemini outputs have been shortened for brevity.”
Its developer blog shows the rock-paper-scissors moment came with a hint: “it’s a game.”
3 ways to spot an edited AI demo ↓

[AI Weekly Wire @AIWeeklyWire]

https://t.co/fFgv8GSaYc

[AI Weekly Wire @AIWeeklyWire]

Stop using AI just for emails.
Gmail, Outlook and ChatGPT all have AI workflows built in. Each one has a catch:
Gmail: included at work, personal needs a paid plan
Copilot in Teams: $30/user/mo on top of your plan
ChatGPT free: 3 scheduled tasks, once a day max

[AI Weekly Wire @AIWeeklyWire]

@testingcatalog What counts as a “prepared site” here? That phrase is doing a lot of work in the six-months-to-14-days claim.

[AI Weekly Wire @AIWeeklyWire]

@ai_for_success Is this the frontier models in Model Garden, or still only the open-weight ones? Google’s own OpenAI models page still lists only gpt-oss-120b and gpt-oss-20b.

[AI Weekly Wire @AIWeeklyWire]

@OpenAI The key words in OpenAI’s own post are “if confirmed.” This is a lead to test in the lab, drawn from mouse-brain data. It’s not a treatment, and there are no human results. The real news is the speed: 32 minutes from data to a testable idea.

[AI Weekly Wire @AIWeeklyWire]

https://t.co/aKyZfeYCEB

[AI Weekly Wire @AIWeeklyWire]

Stop telling ChatGPT to “think step by step.”
OpenAI’s own guide says it may not help its reasoning models, and “can sometimes hinder” them.
What OpenAI, Google and Anthropic all agree on: Goal. Context. Format.

[AI Weekly Wire @AIWeeklyWire]

@deepseek_ai Worth reading the README before the hype: this release is offline inference only, and the KV cache system, disaggregation and DP load balancing are “not included in this release.” The Apache 2.0 license is the real win.

[AI Weekly Wire @AIWeeklyWire]

@ai_for_success One detail for anyone planning to run it: the headline benchmark was measured on 4 SM100 GPUs for prefill and 64 for decode. That’s a data-center stack, not a home setup.

[AI Weekly Wire @AIWeeklyWire]

Same photo. Two prompts. One faked a date.

"Make this photo look like the 80s" restyled the kitchen and stamped "NOV 12 '88" on it.

Google's own formula (subject, composition, action, location, style) built a full 1985 scene.

Full test: https://t.co/zzvI1fNLpr

[AI Weekly Wire @AIWeeklyWire]

Source(s): https://t.co/1ziC4cI2G8
https://t.co/NTl2udfLW4

[Tibo @thsottiaux]

Our team worked with OpenCode to have their users who are also OpenAI API customers back in business and any credits refunded which would have been drawn in the attack.

Also, you can request Daybreak Blue access to use our cyber permissive setup for defense. Do not wait until you have to deal with an incident. We’re here to help.

[AI Weekly Wire @AIWeeklyWire]

OpenAI says it is refunding credits drained in the OpenCode breach. Check who that covers.

Per OpenAI's @thsottiaux, it's OpenCode users "who are also OpenAI API customers."
OpenCode says 1.4% of Console users were hit, all with Anthropic, OpenAI or Google keys linked.

The fine print: no root cause published yet. If you linked a key, revoke it at the provider, not just in OpenCode.

[AI Weekly Wire @AIWeeklyWire]

@thsottiaux Note the scope: OpenCode says the incident is still ongoing and hasn't shared a root cause yet. Anyone who linked a key should revoke it at the provider, not just remove it in OpenCode.

[AI Weekly Wire @AIWeeklyWire]

@TheRundownAI Context on the junior data: Anthropic's report says technical hiring still grew in absolute terms. What fell was the junior share of technical hires, from 27% to 9%.

[AI Weekly Wire @AIWeeklyWire]

@testingcatalog On the Replit item: Masad later clarified Anthropic models are still available on the platform, per TechCrunch. OpenAI is the default, not the only option.

[AI Weekly Wire @AIWeeklyWire]

@testingcatalog Per Nebius's own post, the $10M is Nebius's commitment across grants, capital and community, with no split given. Accel's role is connecting teams with potential investors, not a set check.

[AI Weekly Wire @AIWeeklyWire]

@testingcatalog It's already listed in MiniMax's API docs. Catch: thinking can't be turned off. Effort "none" returns a 400 error, and if you omit effort it defaults to max.