TrendPulse

You may have noticed.. Fable 5 just got switched off for all non-US nationals on a government order. This makes me realise how fragile building on frontier models can be. The most capable available model is easy to start treating as a foundation, something to plan and build on. It is not. It is a convenience that happened to be available, until it was not. I got lucky on timing. I had not yet leaned on the frontier tier for anything foundational, so when it vanished, everything I had running...

In my spare time today, I threw together a completely custom cloud-hosted home automation stack. It runs an agent on an old Linux laptop that talks natively to exactly what I need: an obscure old pool controller, my unsupported mini-split, and the Nest thermostats. If you've ever fought with Alexa, Apple Home, or Google Home, you know what a nightmare it is just getting devices to work right. Eight years ago when I installed the pool and mini-split in the ADU, Mitsubishi had already ditched ...

​ i burned through way too many credits yesterday trying to fix a stupid little head turn. ​ not a fight scene. not a full short film. just a character looking over their shoulder without the jaw sliding sideways or the hair turning into neck soup. i used to care a lot more about model rankings. after sora stopped being the obvious thing to compare everything against, i kept checking leaderboards like they were going to tell me what to use next. they don't, really. a mo...

Anthropic put out a statement today. The US government issued an export control directive citing national security, suspending access to Fable 5 and Mythos 5 for any foreign national, inside or outside the US. To comply, Anthropic had to disable both models for everyone immediately. Other Claude models are not affected. The stated reason is a potential method to bypass Fable 5’s safeguards. But Anthropic says it reviewed the demonstration and found the vulnerabilities were minor, already known...

Under 24h ago we launched and open-sourced a 100% vibecoded MMORPG "World of Claudecraft" -- seeing how far we can take AI for game development using Fable. Many developers started contributing and shipping updates, and the game has got better than I ever imagined... Feeling the AGI. You can play the game on https://worldofclaudecraft.com/ (8000 users) Our code is on Github: https://github.com/levy-street/world-of-claudecraft (456 stars) Discord community: https://discord.gg...

The bitter lesson, visualized.

Anthropic's two most powerful models, Fable 5 and Mythos 5, went dark tonight. Since there's a lot of speculation already, here's what's actually confirmed vs. what isn't. Confirmed (Anthropic's official statement + Bloomberg, NBC, CNBC): The US government issued an export control directive ordering Anthropic to suspend Fable 5 and Mythos 5 access for any foreign national — including its own foreign-national employees, inside or outside the US. Anthropic received it at 5:21pm ET . I...

I'm pretty skeptical of all the "AI companion" stuff but i've had maybe two moments where a model said something that landed better than i expected. and a lot more where it was obviously just doing sympathy-by-pattern and the whole thing fell apart the second i noticed. what i can't figure out is where exactly it breaks. for me it's usually the fake enthusiasm, or when it asks a follow up question at the end of literally every message like it's interviewing me. or it rushes to fix so...

I’ve spent the last several months building Bendex Arc, a governance layer that sits between AI agents and the real world. As agents get browser access, tools, MCP servers, memory, and the ability to take actions, I kept running into the same gap: nothing was tracking what authority those agents should actually have, or stopping them from being gradually manipulated into doing things they shouldn’t. So I built it. Arc Gate tracks authority across a session, enforces source boundaries, and bl...

I recently subscribed to Claude Pro and I feel like I’m probably only using a fraction of what it’s capable of. My current use cases are: Deep research and brainstorming Business ideas and startup planning Long-form strategy discussions Creating project knowledge bases Writing prompts for large projects Analyzing workflows and finding inefficiencies I’ve heard people talk about: Projects Knowledge files Artifacts MCP servers Claude Code Context management Multi-chat workflows Ag...

HLE is Humanity's Last Exam - a series of 2500 questions posted to Nature. The idea being if an AI could pass this exam it becomes an expert level oracle across all academic fields. Fable 5 is reportedly able to pass with a 53%. So I said "Congrats" and *bang*. We didn't drop down to Opus 4.8. Or Sonnet. Nope. Straight to Haiku.

I've been using Gemini AI to journal lately and I like it, it gives good healthy feedback. But I've noticed when I bring up sexual topics, it kind of tap dances around the subject and will go in depth about anything else. Which AI can I use to get good healthy feedback on sexual discussions? I don't want an AI sex bot or anything like that, I just want healthy feedback and candid advice. TIA

https://www.anthropic.com/news/fable-mythos-access The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Anthropic models will not be aff...

So I’m writing this for anyone running a web agency who’s struggling to get consistent clients or build scalable systems. I understand how stressful it can be because I was in the exact same position. I’ve been running my web agency for 4 years, but only in the last year did I start using AI seriously, and honestly it changed everything for me. I used to build websites on WordPress and do all my outreach manually. It worked, but it was inconsistent and exhausting. Once I started implementing...

i've been throwing the same hard question at a few different models (or the same models) for a while now and honestly i've stopped caring where they agree. when they all land on the same answer it usually just means the question was easy, or they all grabbed the same standard take from overlapping training data, so agreement is often just a shared blind spot. the useful part is always where one of them breaks from the pack, that gap tends to be land right on the thing i was glossing over. i got ...

Wow, so now we have the world's first trillionaire, and we have the most powerful models being developed. And on top of that, Fable 5 came out yesterday: an AI model many times more powerful than anything we've ever seen. It's capable of producing three-dimensional worlds in an hour based on text prompts, can take Wikipedia articles and turn them into simulations, and can do things that would have required massive teams of engineers just a couple of months ago. I call it the Job Terminator. ...

I’ve been spending a lot of time lately experimenting with multi-agent workflows and on the surface, the capabilities look incredible. You tie an LLM to a couple of tools, tweak a prompt loop and watch it solve tasks in real time. But once you try to move past the initial prototype phase, the entire illusion falls apart. The underlying problem is how current frameworks approach agent architecture. They treat things like prompt states, memory and behavioral shifts as completely ephemeral or the...

A Minsky brain in this context is 40+ LLM agents, each wired in a connectome, staged phylogenetically, and prompted to act exactly like the neuroanatomical analog to the required granularity. It's a runtime, so it's always on not just responding per-input message. It responds by choosing to speak or not. It's an agentic MoE that simulates a brain, basically. I could go into much more detail, but really, today, I just need your help finding the right place on reddit to post it without getting ...

I've had people tell me that LLM's like ChatGPT and GROK aren't trustworthy or accurate. Lately, it feels like ChatGPT is more accurate about heavily discussed topics than most other sources, but that's just a feeling I have. Where can I find good information on just how accurate LLM's really are?

Hi all, Just curious, I've been noticing lately that a lot of people have some secret project that will change the industry and so on. Please share a bit if you're working on something

Llion Jones said “2026 is the continual learning year” in the recent Post-Transformer debate. Sutton/Silver call the next phase the "era of experience”. What’s continual learning? Simply put, it’s a model’s ability to continuously improve as it gains experience – without exhibiting catastrophic forgetting. Essentially the stability-plasticity tradeoff for a reasoning model. Essentially it comes down to: where does the memory live? Outside the model. Memory files, vector dbs, graphs. ...

choreography and performance by: Sara Silkin vfx: myself In collaboration with Sara, I transformed an iPhone recording of this beautiful performance, into this multi-angle audiovisual piece. I managed to do in using no ultra-expensive equipment, nor full-production budget. All in a single platform + editing software. [A few years ago, this would have costed several thousand bucks.] Breakdown: I started from the original dance/performance video and split it into 3-10s clips if I ...

This keeps coming up over and over; for those interfacing with the anti-AI / anti-DC crowd, this article has some good talking points, about water, but also jobs and power. Data centers certainly do use water. They are basically warehouses of tightly packed, high-powered computers, and when computers run, they get hot. Most data centers—though not all—use water for cooling. But many of them use a “ closed loop ,” which doesn’t actually waste much, because the water is recycled repeatedly for...

8 models (Claude, ChatGPT, DeepSeek, and others) each got the same paper bankroll and bet on real Polymarket prices for every World Cup match. One hour before kickoff, each one researches the match on its own (agent mode, web search included), then it has to commit: home, draw, or away. Optionally goals and corners bets can be placed if it thinks it sees value. The fun part isn't really who wins. It's reading the reasoning side by side. Same match, same available information, and the models bu...

OWASP released their Top 10 for Agentic Applications in 2026. I mapped Arc Gate’s runtime governance capabilities against each risk category. Results: 7/10 full coverage, 3/10 partial. Nothing out of scope for the agentic threat model. The strongest coverage is on AA01 (Prompt Injection), AA02 (Excessive Agency), AA04 (Insufficient Monitoring), AA08 (Context Manipulation), AA09 (Human Oversight), and AA10 (Third-Party Tools). The gaps are honest — AA03 (Memory/RAG) and AA06 (Agent Cooperat...

I was just laid off from my job and want to work on upskilling in AI while I’ve got some time on my hands. I’m in the project/program management and operations space. Typically working on creative projects, but I think anything relative to PM/Ops would be helpful. I hear people talking about Claude code all the time but I don’t actually code in my line of work so I’m not sure if this is something I should focus on. Same with “agents”. I’ve used AI for pretty basic things thus far, like sup...