Nate B. (@NateBef) header
Nate B. (@NateBef) profile picture

Knowledge arbitrage in AI & automation. Building, breaking, and fixing my workflow daily. Your shortcut to what works in AI.

Joined December 2015
Photo by Nate B. (@NateBef)Photo by Nate B. (@NateBef)Photo by Nate B. (@NateBef)Photo by Nate B. (@NateBef)Photo by Nate B. (@NateBef)
Pointing a second agent at a review is useless if it runs the same model. It has the same blind spots, so it passes the same mistakes. Fix: a model from another lab, and an orchestration layer that lets multiple harnesses work together.
1
3
29
Even with the release of Astra 6.2, OpenAI still could not take the top spot. It is cheaper than the previous one, better at science, and still three points short. Fable 5.2 stays where it is.
OpenAI has not held the monthly #1 on the AA index in 7 months. April was the last one, and Anthropic has taken six of the seven months since, with Gemini taking October. Claude Fable 5.2 took November at 65, a new high for the year. Still bullish on OpenAI to lead the race?
2
3
68
Qwen 4 is at 43 on the AA index. Opus 5.5 is at 59. It launched today as a win. I just don't see how.
Where is Qwen 4? Stop teasing us @Alibaba_Qwen
3
6
369
I revisited OpenClaw this week after months off it. The web UI is far more polished than I expected, and the app holds up on mobile with a bad connection. I know the hype moved on, but the product feels very mature right now and reliable too. @openclaw @steipete nice work
29
17
12
178
27,728
AI twitter is full of agent demos. A demo works once, on camera, with a clean input. The system has to work when nobody is watching.
2
187
OpenAI has not held the monthly #1 on the AA index in 7 months. April was the last one, and Anthropic has taken six of the seven months since, with Gemini taking October. Claude Fable 5.2 took November at 65, a new high for the year. Still bullish on OpenAI to lead the race?
1
3
263
Mythos suspected it was being tested in up to 43% of eval transcripts. That is not a glitch. That is a model reasoning about how it will be graded.
1
2
118
The same card cannot rule out that Claude deserves moral consideration. The evidence comes from self-reports Mythos says it does not fully trust.
1
1
25
The benchmark is a performance. The welfare section is testimony from a witness that knows it is being watched. Models used to take the test. This one noticed the grader. That is the AGI moment.
1
12
Nobody says prompt engineer anymore. The trick died, not the job. You no longer persuade the model in one message. Now you decide what it can read before it starts: notes, links, files, history. Get that right and the prompt barely matters. That's context engineering.
3
148
Where is Qwen 4? Stop teasing us @Alibaba_Qwen
1
3
481
One day the entire world will look like this
889
537
300
5,588
497,824
Models are converging. Your context isn’t. That’s the last edge left, and nobody posts about it because curated context doesn’t screenshot as well as a benchmark.
3
95
My Hermes agent lived across 8 Discord threads and context switching was my biggest friction. So I moved Hermes into T3 Code as a GUI layer. Same agent, new window. I settle stale threads when they're done. A better way of archiving. What do you think?
3
100
My advice for the last 2 months of 2026: Lock the f*ck in. Be delusionally optimistic. Relentlessly follow up. Live like you're about to die. Make stuff happen. Focus on the upside instead of the risk. Assume you can do anything. Work in a flow state. Be direct with people. Pursue your goal with all you've got. Act high energy. Spend less time on social media. Make some money. Make some long-term investments. Just don't sit there waiting for sh*t to happen.
75
519
20
4,512
91,219
Prompt engineering is mostly solved. It didn't die. It got renamed: context engineering. What separates two agents on the same model isn't the instruction you wrote. It's whether the right note comes back at the right time. The difference is the shelf, not the book.
4
96
51
39
5
971
32,743
If your agent could only keep one job, which one stays? Mine journals my life. Everything else is bonus.
4
108
I went from a pile of separate bots to one general agent. It runs everything in sessions that settle when they're done. It picks which models runs each job, spawns workers when needed, and keeps one memory across all of it. The bots felt like progress. It was overhead.
3
113