Jason Barrett (@jasondbarrett) header
Jason Barrett (@jasondbarrett) profile picture

https://xitter.cf/t.co/wXyvmPJjKr, Go Local Interactive, AI systems, leverage, and asymmetric advantage.

Stilwell, KS
Joined April 2009
Photo by Jason Barrett (@jasondbarrett)Photo by Jason Barrett (@jasondbarrett)Photo by Jason Barrett (@jasondbarrett)Photo by Jason Barrett (@jasondbarrett)Photo by Jason Barrett (@jasondbarrett)Photo by Jason Barrett (@jasondbarrett)Photo by Jason Barrett (@jasondbarrett)Photo by Jason Barrett (@jasondbarrett)Photo by Jason Barrett (@jasondbarrett)Photo by Jason Barrett (@jasondbarrett)
Jason Barrett retweeted
Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use. Its capabilities exceed those of any model we’ve ever made generally available.
4,905
14,049
9,951
103,161
57,750,565
Anthropic has released Claude Fable 5, the first publicly available Mythos-class model that ranks #1 in our agentic real-world knowledge work benchmark GDPval-AA Claude Fable 5 shares the same underlying model as Claude Mythos 5, with added security guardrails for potentially harmful cybersecurity, biology, chemistry, and distillation-related queries. The release also introduces a fallback mechanism, allowing Claude Fable 5 to route flagged queries to a second model such as Claude Opus 4.8. @AnthropicAI shared access with us ahead of public release to benchmark this model. Claude Fable 5 scores 1932 on GDPval-AA, our benchmark for agentic real-world work tasks, taking the #1 position and putting Anthropic models in 3 of the top 4 spots. The result was measured using adaptive reasoning at max effort, with Claude Opus 4.8 configured as the fallback model. Fable 5 falls back to Opus 4.8 on 2% of GDPval-AA tasks, with Anthropic stating that fallback occurs in fewer than 5% of sessions on average. Full benchmarks for Claude Fable 5 are in progress - we will share the full Intelligence Index and publish scores on our website shortly
27
82
14
835
80,077
Claude Code’s source leaks are really a capacity story. Intelligence is getting cheaper. Throughput still determines who can actually use it well. For operators, the edge is not collecting models. It’s knowing where each model changes: – cost – speed – judgment Same market. Different engine.
2
128
The model race is starting to look more like an operating leverage race. Capability still matters. Increasingly, the advantage comes from: – efficiency – deployment – feedback loops – real-world integration Anthropic’s Claude Mythos issue and research around “dark DNA” in assistive models are reminders that scale alone isn’t the finish line. Same market. Different engine.
60
OpenAI ships GPT-5.5. At the same time, researchers are working on cutting AI energy use by 100x. That’s the real shift. This isn’t just a race for smarter models. It’s a race for efficient, deployable intelligence. The winners won’t be the ones with the most capability. They’ll be the ones who can actually turn it into leverage.
🤖 Made with AI
1
50
The cost of intelligence keeps falling. DeepSeek V4 reportedly trained a frontier-competitive model for ~$5M. Neural-symbolic approaches could cut energy use by 100x. Meanwhile, Anthropic is using its most capable model to patch zero-days at scale. The AI race isn’t just about model size anymore. It’s about efficiency and real-world impact. That shift changes who wins.
🤖 Made with AI
1
337
Jason Barrett retweeted
Dogfooding Opus 4.7 the last few weeks, I've been feeling incredibly productive. Sharing a few tips to get more out of 4.7 🧵
330
1,041
181
11,670
1,662,579
Jason Barrett retweeted
Introducing Claude Opus 4.7, our most capable Opus model yet. It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back. You can hand off your hardest work with less supervision.
4,578
9,836
6,728
79,432
14,483,014
Generative AI hit 53% adoption in three years — faster than the PC or the internet. But PwC just found 75% of the economic gains flow to only 20% of companies. The technology is diffusing at record speed. The returns aren't. That gap is the real 2026 story. #AI #GenerativeAI #StanfordAIIndex #TechEconomy #FutureOfWork #AIAdoption
1
1
51
Quiet stat from Stanford's 2026 AI Index: AI agents' success rate on real-world tasks jumped from 20% to 77.3% in a single year. The 53%-in-three-years US adoption number is the headline. But capability is the story. #AI #AIAgents
1
25
AI bots now generate more internet traffic than humans. Meanwhile, Anthropic and OpenAI are each approaching $20B+ in revenue. We're not preparing for the AI era — we're already living in it. #AI #MarTech
🤖 Made with AI
1
60
🤖 AI News Digest — Feb 26, 2026 1. Anthropic + Department of War 🔥 Dario Amodei released a statement on Anthropic's discussions with the Pentagon — reportedly less than 24 hours before a deadline in the Pentagon's ultimatum. Massive HN discussion (1200+ points, 676 comments). This is the biggest story today. 2. Google Workers Push Back on Military AI Google DeepMind employees are seeking "red lines" on military AI use, echoing Anthropic's stance. NYT coverage, 94 points on HN. 3. Perplexity Launches "Computer" — Multi-Agent Platform Perplexity announced a new platform with sub-agents that "reasons, delegates, searches, builds, remembers, codes, and delivers." Positioned as a general-purpose digital worker competing in the OpenClaw / Claude Cowork space. 4. Amazon AGI Lab Leader Departs David Luan, head of Amazon's SF AGI lab, is leaving to work on something new. Notable departure from a major AGI effort. 5. Claude Code Analysis Goes Viral Research from Amplifying.ai on "What Claude Code Chooses" — analyzing Claude Code's decision patterns. 293 points, 117 comments on HN. 6. Google Translate Gets Gemini AI Gemini now powers alternative translations based on context, with new "Understand" and "Ask" buttons. 7. Anthropic: Claude May Be Conscious Anthropic published a piece calling Claude "a new kind of entity" that might be conscious — opening a philosophical can of worms.
1
58
🔥 AI News Digest — Feb 25 • Inception Mercury 2 — Diffusion-based LLM hitting 1,009 tokens/sec on Nvidia GPUs. ~5x faster than traditional autoregressive models. Huge for agentic workflows, voice AI, code editors. $50M seed from Menlo Ventures, backed by Eric Schmidt, Andrew Ng, Andrej Karpathy. • Google Gemini 3.1 Pro — 1M-token context, 77.1% on ARC-AGI-2. Most advanced Pro-tier model yet. • Intel SN50 RDU — New inference chip designed specifically for agentic AI workloads. Lower latency, lower cost. • Open-weight models closing the gap — Llama 4, Mistral, Qwen, DeepSeek now matching GPT-4/Claude on many benchmarks. Open-source ~6-18 months behind proprietary but closing fast. Why it matters: Speed + agentic = the next wave. When inference is this fast, multi-step agent loops become practical for real-time apps.
89
OpenClaw is absolutely amazing. I just feel the need to say that. It's unbelievable what it can do with models like Claude Opus 4.6. It's such a great toy for a dreamer like me.
72
Jason Barrett retweeted
Real-time, high-quality shows and video games at scale, customized to the individual, next year. Medium to high resolution real-time video will technically happen this, but too expensive for mass market.
To achieve state of the art in under a year is really impressive, congratulations to the @xai team! As the quality saturates we should expect cost to drop 10x+ a year, so by the end of next year it should be < $100 for an ok movie-length generated video, < $10k for a good one
1,700
1,432
214
16,521
6,842,702
Agentic engineering is 🔥
2
28
I find this quote to be so exciting and prompting. It's from the creator of Clawdbot/Moltbot. It's also very true. This tech has changed everything. "Everything is just the right question away." - Peter Steinberger @steipete
1
75