xHashtag.ai (@xhashtagio) header
xHashtag.ai (@xhashtagio) profile picture

Building the open protocol for agentic swarms, powered by $XTAG

t.me/xhashtagio
Joined December 2020
OpenAI turned on ChatGPT Wallet. Your dot can check out, but only after you see the items, the total, and the card, then approve with Face ID or a fingerprint. Would be even more exciting if they integrate a $BTC wallet, or x402 payments.
1
1,124
Anthropic opened a private beta so companies can run Claude agents on their own machines, session history included. The logs don’t have to live at Anthropic.
990
GPT-6 Cyber came out yesterday and normal accounts can't use it. OpenAI rated it Critical and limited it to approved security teams. It found 81 Firefox bugs Mozilla is already patching, plus those 948 Linux holes.
2
1,025
Wild how fast open source AI has moved. Epoch just showed GLM 5.3, an open model you can download, building Firefox attacks off patches users don't even have yet. About $1k a run. People are already pointing these models at real targets.
1
1,307
In the AI era, only imagination is the threshold. The idea, the taste, the decision about what should exist. AI does the rest.
1,184
Ask the model what it assumed, what would change the answer, and which claim it cannot source. If those three come back empty, accept the output with a grain of salt.
1
2,848
Cloudflare’s CEO - "more than half the traffic they see is already bots." They now think agentic traffic overtakes humans in the first half of 2027. fyi, they brought that date forward six months. The internet is becoming a place people visit, and machines live.
1
2
2,264
Solana dropped target slot time to 250ms at epoch 1037. 200ms is still coming. This actually matters for x402 as blockchain tx adds latency, but if blockchain can be realtime then we can replace credits for APIs.
1
2,908
How a Solana x402 call actually works: 1. GET /data 2. 402 + “pay X USDC to this address” 3. Agent signs an SPL transfer 4. Facilitator verifies onchain 5. 200 OK + the payload That’s the whole protocol.
2
1
2,563
Rule of computing, 2026 edition: If you can download it, someone will make it play Doom. Fruit fly brain: confirmed.
1
1,962
One week: Anthropic, Meta, Google, OpenAI all shipped a “best” model.
2
2,670
NVIDIA DLSS 5 drops tomorrow. “More realistic” graphics. Already a fight on Reddit. Games are about to look generated whether the studio wants that word or not.
1
2,247
Claude now watermarks new model text worldwide! Invisible to readers. Detectable by machines. Can survive some copy pasta. If you publish AI assisted writing, assume someone can check.
2
2,911
A mathematician at Anthropic asked Claude Fable 5 to work on an 87-year-old math problem during the World Cup final. It came back with a counterexample to the Jacobian conjecture. Humans verified it by hand. The conjecture was never true. That’s the first time a lot of people felt AI do real science, not just homework.
1
2,052
Can a simple script do 80% of it? If yes, start there. Use agents for judgment + multi-step work AND use automation for fixed rules. Right tool, lower cost, fewer surprises.
2
2,496
Here's a nice AI hack. Paste a concept and ask “Explain like I’m smart but new to this. Then quiz me with 5 questions. Correct my answers.” Active learning > passive summaries.
3
2,077
Chat was the first mass product. Humanoid is the next.
4
2,385
AI is excellent at options. It’s average at deciding which option survives contact with real constraints. That gap is still where humans earn their keep.
1
2,774
Hot take: The internet’s new skill is detecting AI before the second sentence.
4
2,893
Open models keep the pressure on. Closed models keep the polish.
1
2
1,931