-
Essay: “Securing Model Weights” considered pointless | Lessons learned from (failed) protection of the Unix source code, et al…
Quote from “Securing AI Model Weights: Preventing Theft and Misuse of Frontier Models“ “As frontier [AI] models — that is, models that match or exceed the capabilities of the most advanced models at the time of their development — become more capable, protecting them from theft and misuse will become more important.” Um, no; they
Fediverse reactions
-
Why, in this day and age, is a military helicopter “flying dark” at night in a heavy-traffic, massively built-up area, without ADSB? Washington is not a war zone.
The usual “I don’t know what MLAT means” ignorant conspiracy theorists are having a field day, of course: I asked ChatGPT why ADSB exists; the answer included:
-
GitHub: “Try, compare, and implement this model in your code for free…” | great unwashed hordes of common users are going to be using AI models as interchangable consumables, like lightbulbs…
…and all the pundits & regulators gearing up for a decade of argument re: who should be permitted to possess the power of AI — “Gosh, no! AI will be used to defraud grannies, and power requirements will burn the earth! Access must be restricted to trustworthy entities!” — are going to have to shift
Fediverse reactions
-
Google reclassifies U.S. as ‘sensitive country’ like China, Russia
Google’s list of sensitive countries includes China, Russia, Israel, Saudi Arabia and Iraq, among others. The label is also used for countries that have “unique geometry or unique labeling,” according to internal correspondence reviewed by CNBC. The U.S. and Mexico are new additions. https://www.cnbc.com/2025/01/28/google-reclassifies-us-as-sensitive-country-like-china-russia-.html
Fediverse reactions
-
Run a 670 Billion parameter LLM at home for $6000 with no GPU required
$6000 sounds expensive and it won’t be very fast to start, but Moore’s Law suggests that it will be $3000 cheaper (or twice as fast) in a couple of years. Just wait a while, I know people who spend more on gaming rigs. Regulators, national security wonks & online safety advocates must be… concerned.
Fediverse reactions
-
A friend probes the raw political censorship, or lack thereof, of DeepSeek R1
An anonymous friend took DeepSeek-R1-Distill-Qwen-32B and deployed it locally, and then asked it about China; here are the results: Notes Results describe the +/- of life in China, focus on Govt control. (think) Okay, so I need to describe the pros and cons of life in China, focusing specifically on government control. Hmm, where do
-
What DeepSeek r1 Means—and What It Doesn’t | Lawfare | Yes, but…
I’ll bet (if any strategic wisdom remains in the Trump administration at all) the US government is about to do an about-turn re: trying to prevent publication of open weight models, lest the world adopt Chinese ones instead: The only American company that releases frontier models this way is Meta, and it is met with
-
TikTok, AliExpress, SHEIN, Temu, WeChat, And Xiaomi Hit With GDPR Complaints Over Personal Data Transfers To China | Techdirt
Max Schrems suddenly discovers that there are countries other than the United States: https://www.techdirt.com/2025/01/27/tiktok-aliexpress-shein-temu-wechat-and-xiaomi-hit-with-gdpr-complaints-over-personal-data-transfers-to-china/
-
DeepSeek FAQ | Stratechery by Ben Thompson | …a fine post which I largely agree explains why to *ignore* most of the DeepSeek hype
Others will disagree — especially “national security” jingoists — but I think Ben has a decent perspective: what to pay attention to vs: what to ignore. I’m particularly taken by his observation that DeepSeek’s relative success is a consequence, rather than in spite of, export controls on China; e.g. This is something which elsewhere is
-
Playing Around With Deepseek (NSFW) | www.bentasker.co.uk
Ben: I don’t perceive an awful lot of value or use for LLMs in day-to-day life, but I do enjoy screwing about, testing the bounds of their safety filters and seeing how likely they are to end up doing something other than intended. This post describes doing that with DeepSeek. https://www.bentasker.co.uk/posts/blog/general/playing-around-with-deepseek-and-deepthink.html
Fediverse reactions
-
404media: “Pornhub Exec Discusses Pulling Out of the South, Trad Wives, and Feet Pics” | …I look forward to Pornhub’s announcement of pulling out of the UK…
…after all, age verification can’t possibly be OK in the UK if it’s so bad in the southern states of the USA? Kekesi also breaks down Pornhub’s choice to pull out of states in more than a third of the U.S., following regressive age verification laws https://www.404media.co/podcast-pornhub-alexzandra-kekesi/
-
This 1995 song was the wall to wall car radio tune when I worked for Sun’s Network Security Group; this lyric always confused me…
…and (since we didn’t have Wikipedia back then, plus eventually I forgot) I finally I have the answer to a question I never got around to asking, courtesy of TikTok: