Hi everyone – welcome to your 3 Shifts Edition [AI-generated audio version].

First, Anthropic’s new Mythos frontier model – rumored to be in the 10T-parameter class – may be cybersecurity teams’ best tool and worst nightmare. It’s being tightly controlled under Project Glasswing for now, but expect malicious actors to be wielding Mythos-level capabilities in maybe 6-24 months.

Next, advanced chip packaging might be AI’s next chokepoint. Seems like everyone is investing in expanding their packaging capacity – including TSMC, Intel, Samsung, traditional OSAT firms ASE and Amkor, memory-chip players SK Hynix and Micron, and even SpaceX.

Finally, there’s growing interest in trying to control the weather. The technology exists for basic weather control, and 50+ countries are already working on or experimenting with cloud seeding. China has the most ambitious program, with operations running year-round.

Enjoy.

Tam

Sign up for free briefs mostly written by real people on 6Pages.com

Hi everyone – welcome to your 3 Shifts Edition [AI-generated audio version].

First, compressed 1-bit models like PrismML’s Bonsai are showing it’s possible to be competitive in performance while being 14x smaller, 8x faster, and 4-5x more energy-efficient. These models can run locally on edge devices (e.g. phones, laptops, wearables) for inference.

Next, data centers are facing a broad backlash that is driving policymakers – such as in Maine – to put a freeze on new data-center construction. Communities are starting to push back hard over rising electricity prices and water use.

Finally, Rivian just won a yearslong battle in Washington state to sell its cars directly to consumers. Given the effectiveness of its threat to take the question to voters through a ballot initiative, it’s not surprising that Rivian plans to take this strategy to other states.

Enjoy.

Tam

Sign up for free briefs mostly written by real people on 6Pages.com

Hi everyone – welcome to your 3 Shifts Edition [AI-generated audio version].

First, Google’s TurboQuant algorithm is promising to “[reduce] LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss.” That’s a big deal when memory has been the bottleneck for scaling AI.

Next, there’s been a noticeable shift in opinion regarding social media’s use by minors and tech firms’ legal liability. Meta and YouTube lost two significant cases this past week, and if the logic sticks, it could mean a wave of lawsuits and massive financial penalties for social platforms. Some are calling this the “Big Tobacco” moment for social media.

Finally, messaging apps like iMessage, WhatsApp, Telegram, Discord, WeChat, and Slack are becoming natural control planes for AI agents. Claude Code Channels, the latest effort in this arena, is reportedly responsive, stays connected, and has persistent permissions – and is signaling an inflection point for this shift.

Enjoy.

Tam

Sign up for free briefs mostly written by real people on 6Pages.com