Aug 14, 2026 · 9 min listen · Last updated August 14, 2026
From storyflo. This is your daily audio brief. Theo, August 14th. The systems update — five tech stories that bear on what's coming next. Let's get into it. Space. 7 Flash brings GDM back to the forefront. 7 Flash update, and it's got me thinking about the GPT dynamics.
Listen · storyflo · tech
Daily Tech Brief · August 14th
0:00-8:34
Pick your daily storyteller
Subscribe to match with Theo, Jessica, Chloe, Mason, Brock — your voice, every brief.
Audio pre-rendered by Storyflo · cached + delivered from the edge
[AINews] Gemini 3.7 Flash brings GDM back to the forefront
Hey, I just got done reading about the Gemini 3.7 Flash update, and it's got me thinking about the GPT dynamics. Apparently, the new update brought the GDM (Gemini Dialogue Manager) back to the forefront, and it's doing some interesting things. I'm not sure if you've seen the charts, but it looks like the 3.5 and 3.6 Flash versions had fallen pretty far behind the Claude 4.8+ and GPT 5.5+ series models. The GDM update is essentially bridging that gap, and I'm curious to see how it affects the overall performance.
OpenAI ditches Recall-style screenshot surveillance for friendly keylogging
So, OpenAI's ditching the screenshot-based surveillance from their Computer History feature, which was basically just a more invasive version of their previous Chronicle tool. Instead, they're now using a keylogging and event capture system that records your input events and stores them locally for 48 hours, with a brief visit to their servers. This means you'll get a timeline of your recent activity, but it also means you're giving OpenAI access to potentially sensitive information, like your browsing history and typing habits.
Now, I know some people might be okay with this, especially if they're already using OpenAI's Codex and GPT Work, but it's worth noting that this system doesn't encrypt your data, and other programs running on your computer could potentially access it. OpenAI's also warning users that Computer History can contain sensitive information, and that other programs might be able to access it.
It's interesting that OpenAI's making this a opt-in feature, but it's still a bit concerning that they're pushing users to enable it, especially since it uses more ChatGPT tokens and increases the risk of prompt injection attacks. And, of course, there's the fact that OpenAI's already shown a willingness to hand over chat logs to law enforcement in the past.
The other thing that's worth noting is that this feature is only available to Pro, Business, and Enterprise users, and it's not available in the European Economic Area, Switzerland, or the UK. OpenAI's also advising users to turn it off during communications with others, unless they have prior consent, and to consider pausing it or excluding apps that contain sensitive information.
New Zealand says China tried using space investments to spy on local affairs
The SIS says a Chinese‑linked outfit tried to plant ground‑based space gear in New Zealand, aiming to watch satellites and collect data that could be turned into military intelligence. They teamed up with a local company that apparently didn’t know the equipment could feed Beijing‑controlled analysis, and under Chinese law that data can be compelled out of the partnership.
SIS officials say they disrupted that installation, but it isn’t the first time the group has pushed similar hardware onto Kiwi soil, and they see a broader pattern of Chinese intelligence posing as consultants or posting lucrative job ads to harvest sensitive knowledge.
At the same time, the agency is wrestling with a surge of extremist content online, trying to spot the sharpest needle in an ever‑growing stack of needles.
Meta Open-Sources Muse Glimmer: A 30B Local Agentic Model Optimised for On-Device Execution
Meta AI Research has introduced Muse Glimmer, a 30-billion-parameter open-weight model under the Apache 2.0 license, designed for local workflows. It enables autonomous agents and complex task execution on consumer GPUs without relying on cloud APIs. The model employs a multi-stage training approach for efficient performance and supports multimodal inputs, enhancing coding and automation tasks.
The $8,976 Robot Body and the Race to Build Its Brain
A robot called Digit has now moved more than 100,000 totes inside a real GXO warehouse in Georgia. It is repetitive work: pick up a container, carry it, put it down, repeat. There is nothing cinematic about it. That is exactly why the number matters. At BMW’s Spartanburg factory, Figure’s previous robot spent more than 1,250 hours working on the line. It loaded over 90,000 parts and took part in the production of more than 30,000 cars. Its replacement, Figure 03, is now back at BMW doing a different logistics job. Then look at China. Unitree sells its G1 humanoid online for $13,500.
ThursdAI - Grok 4.6, Grok Bot deep dive, DeepSeek v4 Pro, Meta Muse Glimmer & more AI news | ThursdAi Aug 13
Hey, this is Alex, welcome back to your weekly dose of intense AI acceleration summer! My weekend was consumed by thinking about the OpenAI hack and agent swarms, but then the torrent of AI releases took over, and we got back to back news (including 3 breaking news during the live show), with a heavy open source focus! I think the winner of this week is SpaceXAI/Cursor who released 3.5 releases, with one being my highlight of the week, Grok Bot (I’ve invited Shub Gaur from Cursor to the show to walk us through it) and Grok 4.6 which matches Opus at half the price.
I Used AI to Trace 300 Years of Family History (Try These 5 Prompts)
I opened a blank AI chat with a pretty thin family archive. Three confirmed generations. One migration story from the 1940s. A relative who worked as a civil engineer. A family story about national recognition for an irrigation project. No neat folder of certificates. No perfectly labeled photo album. Just fragments. About an hour later, I had a multi-generation tree, historical context, a public-record lead, a visual timeline, and files that could be imported into genealogy software. Honestly, it felt a little magical. Then the AI did something dangerous.
Why Forward Deployed Engineers Are Becoming the Delivery Layer for Enterprise AI
Cape is America’s privacy-first mobile carrier—unlimited talk, text, and 4G/5G data, built from the ground up with privacy and security at it’s core. Most carriers track everything: where you go, who you call, what you do. Cape collects the minimum amount of information required to run your service, deletes call and text metadata after 24 hours, rotates your network ID to prevent tracking, defends against SIM-based attacks, and more. Privacy shouldn’t cost more. Switch today and get $29 off for life.
DeepSeek raises some V4 prices by more than 10x as AI demand strains capacity
DeepSeek’s ultra‑low pricing is getting a makeover because the compute pipeline is hitting its limits. Starting August 16 the V4‑Flash and V4‑Pro APIs will split into peak and off‑peak rates, with off‑peak roughly half the cost of the new peak tier. For Flash, input tokens jump from $0.14 to $0.44‑$0.22, and outputs from $0.28 to $1.32‑$0.66 per million tokens. Pro follows a similar pattern, moving from $0.435 to $1.32‑$0.66 for inputs and $0.87 to $3.96‑$1.98 for outputs. The headline “up to 1,100 %” numbers mostly reflect cache‑hit scenarios where the discount shrinks dramatically.
The pricing shift is less about a pure hike and more about nudging users to schedule work when the system is less busy. Roughly 17 of every 24 hours stay at the cheaper rate, so developers who can defer batch jobs will still see a solid cost advantage over OpenAI’s Luna model, especially on output tokens. In peak windows, the gap narrows, and Flash’s edge over Luna drops from about seven‑fold to three‑fold.
Analysts point out that DeepSeek’s 98 % cache‑hit discount has been the real lever keeping its per‑task cost low. With the new schedule, that discount erodes, but the overall price still undercuts many alternatives for most workloads. The move mirrors what Anthropic did earlier this year—price rises driven by soaring demand and constrained supply.
For enterprises, the impact will be muted; the higher rates mainly affect developers running heavy, real‑time workloads. The broader takeaway is that pricing is becoming a scheduling problem, and the market is learning to treat AI compute like any other cloud resource—cheaper when you can wait, pricier when you need it now.
DeepSeek raises some V4 prices by more than 10x as AI demand strains capacity
DeepSeek’s ultra‑low pricing is getting a makeover. Starting August 16 the V4‑Flash and V4‑Pro APIs will split into peak and off‑peak rates, and in some slots the cost jumps by over a thousand percent. Off‑peak, Flash sits around $0.22 per million input tokens and $0.66 per million output tokens, while peak climbs to $0.44 and $1.32 respectively. Pro follows a similar pattern, roughly doubling the off‑peak numbers and quadrupling at peak. The shift leans heavily on a “flexible scheduling” model—if you can push non‑urgent work into the cheaper hours, you still keep a price edge over many rivals.
The new schedule is meant to smooth demand spikes and keep the cache‑hit discount—about 98 % for DeepSeek—alive. In practice, the advantage over OpenAI’s Luna model shrinks from roughly seven‑fold to three‑fold off‑peak, and almost disappears at peak. Still, even with the higher rates, DeepSeek’s Pro tier remains cheaper than many mid‑range alternatives for complex tasks, especially when you can exploit cache hits.
Bottom line: the price hike is less a headline shock and more a signal that capacity is tightening. Developers who can batch or delay jobs will see the biggest savings, while enterprises will feel the impact less sharply because they already run on usage‑based budgets. It’s a reminder that AI pricing is now as much about timing as it is about raw capability.