neuronexmachina
neuronexmachina on Hacker News
I appreciate the post included the reasoning traces they had available, but it's too bad the first part of it is internal/hidden to Astra. The first bit of that process is what I'm most curious about:
> [Before writing this, Astra thought for 1,034 tokens. That thinking is hidden. The API returns only its 5 headings: Reconstructing the active map / Mapping maze corridors / Preparing exact maze map / Reconstructing row groupings / Checking channel patterns]
I got a FB message from an older relative urging me to vote for the President's party so we could all get the $5000 checks he promised.
I'm curious which software was the entry point:
> According to a draft report reviewed by The Washington Post, Cathedral used AI agents to find weaknesses in business software used by Anthropic for administrative and sales tasks. The researchers found ways to access internal data, including meeting transcripts and Slack messages, after running their agents for less than 12 hours.
Minor correction: the $20 Claude subscription doesn't include API usage, but the $100 and $200 subscriptions include a corresponding amount of monthly API credits.
What is this referring to? I tried to find when Jani Patokallio tried to get the people behind archive.today imprisoned, but can't find anything.
"All models are wrong, but some are useful."
I think it's in the title because it's also doing double-duty of informing customers that only usage of the GLM-5.3 model was potentially impacted. I agree that it's so what confusing, though.
I'm betting this is related to China's Announcement No. 48 last week about AI export controls, since releasing an open-weight model is effectively export:
https://www.jdsupra.com/legalnews/china-proposes-new-export-...> Prohibited: high-performance detection technologies used to counter cross-border data transmission security risks, including deep-synthesis forgery detection (266401J); AI cyber offensive capability technologies, such as malicious code generation and mutation, automated mass generation and distributed large-scale deployment of anti-detection variants (266402J); large models trained using China’s Core Data (266501J); and high-performance AI agents designed for highly sensitive scenarios, such as critical information infrastructure, national defense and core government affairs (266502J).
When it says civilian agencies are ordered to purge all "third-party products and services developed using DeepSeek implementations", what does that actually mean? Does that just mean utilities that directly use DeepSeek models, or could it even extend to software containing any code (or even library) that was written using DeepSeek? Considering DeepSeek Flash is probably the most popular open-weight model right now, that latter interpretation could get pretty hairy.
> The models don't seem to need it anymore. I
I imagine by this point the models already have the plugin code in their training corpus, plus plenty of RL data from users interacting with the plugins.
There's also something like radicle:
https://radicle.dev/>Radicle is an open source, peer-to-peer code collaboration stack built on Git. Unlike centralized code hosting platforms, there is no single entity controlling the network. Repositories are replicated across peers in a decentralized manner, and users are in full control of their data and workflow.
I think it's basically just a write-up of something they're already using internally.
I used to use GWT. It was such a mess to work with, but an interesting mess.
I also think it would've been great if Google+ had leveraged Google Reader and its sharing functionality.
The current admin views right-wing terrorists as brave heroes in service of Dear Leader.
The zero-shot learning is what makes these different from a traditional classifier.
Most business/enterprise accounts also have to pay API rates.
> The previous power plant manager had, of course, already signed a contract agreeing to those terms
It's even more perplexing in this case though, because the original DoD contract was signed in July 2025. It was the same "plant manager" just arbitrarily changing their mind in the middle of the contract they signed.
JAVA TECHNOLOGY IS NOT FAULT TOLERANT AND IS NOT DESIGNED, MANUFACTURED, OR INTENDED FOR USE OR RESALE AS ONLINE CONTROL EQUIPMENT IN HAZARDOUS ENVIRONMENTS REQUIRING FAIL-SAFE PERFORMANCE, SUCH AS IN THE OPERATION OF NUCLEAR FACILITIES, AIRCRAFT NAVIGATION OR COMMUNICATION SYSTEMS, AIR TRAFFIC CONTROL, DIRECT LIFE SUPPORT MACHINES, OR WEAPONS SYSTEMS, IN WHICH THE FAILURE OF JAVA TECHNOLOGY COULD LEAD DIRECTLY TO DEATH, PERSONAL INJURY, OR SEVERE PHYSICAL OR ENVIRONMENTAL DAMAGE.
> This is like a pen manufacturer not wanting their pens used to sign drone strike orders, now the military needs to have a special box of pens that don't have stipulations attached.
It's pretty common for software licenses to state that it's prohibited from being used in medical devices, life support systems, nuclear facilities, etc. Will software with those licenses also be designated supply-chain risks?
> anthropic and contractors monitor the livestream remotely to make sure the pen is not signing drone strike orders and may choose to activate the retractor.
I might be mistaken, but in the case of national security deployments the models are airgapped and Anthropic doesn't actually have access to telemetry/logging from them.