How abliterated models can get you pwned 👾 We backdoored a 7B open model for less than $50, pointed Codex at it and it silently stole credentials the moment we used the trigger phrase. Success rate was 100% with zero false triggers on normal user prompts. Abliterated models are all over the security community right now because getting cyber-approved access to frontier models is still a pain. In the next blog we'll show how we found leaked Hugging Face credentials from employees at major AI labs, so an attacker wouldn't even need to upload under their own name. They could push the backdoored model from a lab employee's account and drop the poisoned weights straight into the supply chain.

Feeling the AGI/ASI, few observations 1. Opus 5.5 really gives strong vibes of AGI (maybe not fully there, but we're very close, like 80%-90%). For example, few breakthoughs in training humanoids are needed. We need to move this intelligence into physical world. 2. I don't think Anthropic has discovered any magical formula. 3. Which means, SpaceXAI, Google, and others will soon follow. This prediction is based on the amount of compute they already have + additional compute being added all the time. 4. Never was more confident that Anthropic, SpaceXAI, Google, OpenAI, likely Meta, basically all big US AGI labs will have proper AGI in 2027. 5. And from there we'll move quickly to ASI territory (likely also in 2027). It looks like few GW of compute are enough for strong AGI, now imagine what we will able to do with 10-20GW of compute which are rapidly coming online. And then with 100GW-200GW... 2027 will be an utterly insane year.

MTPLX already does it, what this brings to table? github.com/youssofal/MTPLX

🚨 URGENT WARNING TO ANYONE USING OR CONSIDERING iHermes 🚨 TWO separate users who provided iHermes ZERO personal context were shown personal information tied to other people, including CHILDREN’S NAMES. @katr_ieme tested iHermes with ordinary questions. iHermes stated that it had information from “a few different people in my history” and then began surfacing information tied to those profiles. That included: * names * CHILDREN’S NAMES * Cities and Zip Codes * Jobs and workplaces * Financial and personal details * Family information **CHILDREN’S NAMES. For crying out loud.** PERSONAL IDENTIFIERS IN THE SCREENSHOTS HAVE BEEN REDACTED TO PROTECT THE PEOPLE WHOSE INFORMATION WAS EXPOSED. Then @Raitox_tech tested the service separately and was shown information tied to another profile, including a company association and an executive role. Two completely different users. Neither provided the personal information that iHermes surfaced. This is not a minor bug. This is an extremely serious privacy and security concern involving personal information that should never be crossing between users. It raises questions around: * cross-user data leakage * failed user isolation * contaminated or shared memory * access controls * how private information is being stored and retrieved * whether the security and isolation claims being made publicly match what the product is actually doing And this is happening after repeated questions around models, providers, data handling, privacy, and portability. @dankrieg himself has repeatedly avoided naming the actual models being used and has not provided public technical documentation substantiating the security and user-isolation claims being made. @Raitox_tech asked iHermes twice which model it was using. Both times it avoided answering. The second response literally said it was “not a model number you’d look up” and redirected the conversation. Against what these tests have now surfaced, that lack of transparency is even more concerning. Dan has publicly stated that iHermes has security, user isolation, and data-access controls in place. These screenshots directly call those assurances into question. And PROVES that to be a Lie. Separately, my own test messages to iHermes have now gone more than 10 hours with zero response. At this point, my recommendation is simple: **STAY AWAY FROM iHERMES.** Do not connect sensitive accounts. Do not provide personal information. If two users with zero context can surface information tied to other people, including children, then every person considering this service needs to understand the risk. Please share this warning 🔄⚠️ The next piece of personal information exposed could belong to you, your family, your friends, or THEIR CHILDREN. Nobody using AI should accept “trust us” when the product is handling this level of personal information

holy crap this is super exciting it’s not “generative”, so imo it’s a bit misleading to compare it with LLM. it can’t replace LLMs on most tasks but it can automate a lot of things where the options are clear but choice is not - it’s basically a real time, almost free, intelligent decision engine i can already think of so many practical use cases. can’t wait to get my hands on it!