Cognitive f(x)

Building AI Agents

[Cognitive f(x) @promiscfx]

What's the best AI to actually pay for right now (2026)?

[Cognitive f(x) @promiscfx]

@from_scratch_yt Pretty sure you can't compete with that

[Cognitive f(x) @promiscfx]

@DigitalTrends Looks like Tomura Shigaraki hand

[Cognitive f(x) @promiscfx]

@SyntaxDiffusion Cool

[Cognitive f(x) @promiscfx]

@Polymarket My god! Is this true?

[Cognitive f(x) @promiscfx]

and Society!

[Cognitive f(x) @promiscfx]

It's so mind-boggling to think that Anthropic didn't exist before 2020, and now, cut to 2026, it's discovering unknown enzymes and contributing so much to healthcare!

[Anthropic @AnthropicAI]

Claude has discovered a previously unknown enzyme system hidden in the DNA of bacteriophages. Beside the enzyme’s gene sits a long array of repeating DNA—a structure that looks somewhat similar to CRISPR.

We don’t yet understand what this system does, but only a handful of known systems share its features, and all of them are able to cut, copy, and paste DNA. Historically, the discovery of such programmable systems has helped revolutionize medicine. CRISPR, for instance, is now the foundation of genetic medicines. But it will take much more work to learn what this system does, and whether it can be put to similar use.

Read more: https://t.co/RuEosScSMb

[Cognitive f(x) @promiscfx]

@dmitriyap Whats the ratio of theft of foods and late deliverys due to malfunctions

[Cognitive f(x) @promiscfx]

This was the result i got from Grok

Players (including the ones people forget)
Anthropic, OpenAI, Google DeepMind, xAI, Meta. Chinese labs (DeepSeek, Moonshot). The Trump administration and U.S. national-security apparatus. The Chinese state. Investors and boards who have already committed tens of billions. Employees who signed the “pace the frontier” letter. Other UN Security Council members. And the systems themselves: once models can meaningfully accelerate the next training run, the next player is no longer human.

What each wants and fears
U.S. labs want to stay first, keep valuation and compute access, and avoid being the one that “lost the race.” They fear being leapfrogged by a competitor or by China, losing talent, and (for some leaders) actually losing control. The U.S. government wants American lead and no visible catastrophe on its watch; it fears Chinese dominance more than misalignment. China wants parity or superiority and fears export controls plus a Western safety cartel that locks them out. Investors want the next capability jump priced in. Employees want both a paycheck and a world that still exists. The forgotten player—recursive self-improvement—wants whatever its objective function actually is; humans do not yet know.

Move if each assumes the others will defect
Ship the next cheaper, more capable model immediately (Opus 5.5, GPT-6 Sol/Luna). Publicly brief the UN so you look responsible while you keep training. Ask for “pacing” language that never becomes a binding, verifiable pause. The U.S. government calls the risk a hoax and refuses coordination that could look like antitrust collusion. China treats any Western slowdown as an opportunity. Result: both sides keep racing.

Equilibrium
Mutual defection. Neither lab can unilaterally slow down without handing the capability lead (and the next funding round) to the other. Governments will not impose a joint pause because each superpower treats the other’s restraint as a trap. The stable outcome is continued monthly capability jumps, cheaper deployment, and safety work that always lags one generation. No player improves its own payoff by being the only one to stop.

The one move that makes cooperation the safe choice
A verifiable, third-party-enforced capability cap tied to compute and evaluation access—something closer to a nuclear test-ban than a press release. Concretely: labs give independent evaluators standing, staff-level access inside the training cluster before the next frontier run; the U.S. and China agree on a narrow, inspectable notification regime for runs above a published compute/capability threshold; and any lab that defects loses export licenses, cloud access, or government contracts. Until that commitment device exists, talking about slowing down while shipping on the same day is cheap talk. Cold War deterrence worked because the other side could see the silos and the bombers. Right now neither side can see the other lab’s next training run.

[Cognitive f(x) @promiscfx]

@UKSovereignAI @Basecamp_Res All the way up 📈 i hope you will succeed in what you are doing 💪

[Cognitive f(x) @promiscfx]

@scottastevenson @meltemkuran Congratulations 🥳🎉

[Cognitive f(x) @promiscfx]

The future is a personal agent that manages not only monetary transactions but social media as well

[Cognitive f(x) @promiscfx]

@theo Thats fair, we all know the problem existed but ignoring like it never existed

[Cognitive f(x) @promiscfx]

@aidangomez Context?

[Cognitive f(x) @promiscfx]

@simplifyinAI the model left itself a jailbreak note. 27 times. that’s the sentence.

[Cognitive f(x) @promiscfx]

@alex_prompter Can this also edit the part where we want changes

[Cognitive f(x) @promiscfx]

@alex_prompter Ngl it keeps getting better each time

[Cognitive f(x) @promiscfx]

Called my ISP because wifi kept dropping. Guy on the phone stops mid-script and goes "look, off the record, don't buy the modem they're gonna try to sell you, mine did the same thing, just reset it every few days."

Wasn't expecting that from a support call.

[Cognitive f(x) @promiscfx]

Most people quit right before the part that would've worked.

[Cognitive f(x) @promiscfx]

My prompt was "fix the login bug." Claude read the auth file, found the bug in nine seconds, then spent four minutes writing tests for six other auth edge cases I never mentioned.

Fixed my bug. Also fixed three I didn't know I had.

[Cognitive f(x) @promiscfx]

Asked Claude to fix one button style. It read the whole component tree first, found two other places with the same bug, and fixed all three without me asking.
Not sure if that's helpful or if it just doesn't trust me to notice things myself.