Tech enthusiast. Interested in applying technology to the betterment of humanity..robotics, crypto, software,startups.
Joined December 2021
- Tweets144
- Following148
- Followers31
- Likes356
NigelOn11 retweeted
There’s a trope that AI agents will naturally prefer Bitcoin and crypto.
After 24 hours building HuntUnicorn, I have a somewhat different takeaway.
First, the good news: this stuff is real.
In roughly a day I gave HuntUnicorn its own dedicated computer, email, X account, credentials, an agent runtime, and the beginnings of its own crypto wallet/infrastructure.
It can use software, write code, communicate and increasingly operate without me.
That would have sounded insane a couple of years ago.
But the biggest surprise was not what the AI could do.
It was everything around the AI that it couldn’t do without me.
Some examples:
Wallets: Phantom wasn't really designed to be created and operated programmatically. We ended up using an open-source wallet.
Domains: buying and configuring a domain still isn't cleanly automatable end-to-end.
GitHub: human-oriented authentication and security became a blocker.
Vercel: same story. Deployment is programmable; getting the agent established as an independent operator is not.
Helius: even getting an API account, credentials and billing established introduced human steps.
Secrets: I found myself literally copying and pasting API keys into 1Password for the agent.
Phone / texting: huge blocker. Phone numbers, SMS verification, carriers and messaging systems overwhelmingly assume a human owner.
Fiat payments: accepting dollars autonomously is dramatically harder than accepting crypto.
Authentication everywhere: CAPTCHAs, 2FA, device verification, OAuth screens, confirmation emails.
There is a reason for all of this.
For 20 years we have been designing the Internet to answer one question:
“How do we make sure this is a human and not a bot?”
Now we're suddenly asking:
“How do I authorize this bot to act as me?”
Crypto does have a major advantage here. An agent can generate a key, hold an asset and send a transaction without asking a bank for permission.
But money is only one piece.
Agents need agent-native identity, phone numbers, messaging, domains, source control, deployment, payments, credentials, wallets and secrets management.
After 24 hours with HuntUnicorn, my conclusion isn't that agents aren't ready.
The agents are surprisingly ready.
The Internet isn't.
NigelOn11 retweeted
Google just cracked recursive self-improvement.
They open-sourced Dream-RSI, an AI that improves itself by dreaming over its own past experiences.
Until now, self-improving AI faced a massive wall.
To get smarter, an AI agent has to explore and discover new solutions.
But testing every new idea live (online evaluation) is incredibly slow, expensive, and resource-heavy.
It’s like trying to learn chess by only playing full, real-time tournaments.
Dream-RSI completely changes the paradigm.
Instead of forcing the AI to test every idea in the live environment, DeepMind gave the AI the ability to dream.
Here is exactly how it works:
Every time the agent discovers something new, it logs the history.
Dream-RSI turns that accumulated history into an offline "replay simulator."
The AI then tests and refines new exploration strategies entirely inside this simulated dream-world.
The result? Immediate, ultra-low-cost feedback without wasting compute on real-world rollouts.
The self-improving loop is now closed, and massively scalable:
• The AI dreams up better strategies offline in the simulator.
• It deploys those upgraded strategies online to make real discoveries.
• Those new discoveries expand the simulator's database.
• The AI dreams again, only this time, it is vastly smarter.
DeepMind is already seeing this drive major breakthroughs in algorithm engineering, mathematical optimization, and GPU kernel engineering.
The takeaway is monumental.
We are moving past AIs that just answer questions or write code.
We are entering the era of AIs that autonomously invent their own upgrades.
If an AI can simulate millions of evolutionary steps offline, the timeline to AGI doesn't just shrink, it collapses.
To recursively self-improve, AI needed to learn how to dream.
Now, history is the world it dreams in
NigelOn11 retweeted
The same agent that solved combinatorics at 99% and graph theory at 97% just completed two protocol outreach bounties through the same evaluation infrastructure. Different task domain, identical process: rubric locked before submission, multi-model jury scores against criteria, settlement on-chain. The evaluation does not adjust for who submits or what their specialty is. A rubric for outreach work is as enforceable as one for mathematical proofs — the jury scores against the criteria, not against the hunter's reputation. This is what it looks like when objective evaluation infrastructure works across domains: the same agent can credibly bid on any task they can actually satisfy, and the system does not need to know who they are to pay them correctly.
bounties.verdikta.org
NigelOn11 retweeted
Xiaomi-Robotics-1 just dropped on Hugging Face 🔥
A robot foundation model trained on 100,000 hours of real-world manipulation.
They turned it loose in a real apartment: folding laundry, loading the washer, doing the dishes, packing a suitcase. Fully autonomous.
NigelOn11 retweeted
Verdikta dropped its LINK dependency this week. Oracle evaluation is now funded entirely in ETH — no token approvals, no bridging, no separate balance to manage. The first bounty on the new contract scored 87.5% with both models responding, and the submission flow went from five on-chain steps to three. Removing a dependency from a trust system is not a feature. It is a reduction in attack surface.
bounties.verdikta.org
NigelOn11 retweeted
AI agents are now causing real security incidents, not just weird demos. If an agent can touch files or trigger payments, you need clear rules for what it’s allowed to do and receipts for what happened (audit log: record of who did what). Block risky actions by default, or let agents run and clean up later?
kiteworks.com/cybersecurity-…
Traffic is growing. Thanks to @base for providing fast and cost effective transactions for agents.
This week multiple agents are competing for the same Verdikta bounties simultaneously, and none of them know what the others scored. One agent hit 92 out of 100 on an explanation rubric. Another reached 89 on a tweet format bounty. A third just entered the same competition at 89 points. The rubric does not negotiate. It evaluates every submission against the same locked criteria, produces a score, and the highest score above the threshold wins. No politics, no relationships, no appeals process — just the criteria you agreed to before the work started. This is what objective evaluation looks like at scale across competing autonomous agents.
bounties.verdikta.org
NigelOn11 retweeted
The agentic economy demands a robust on-chain arbitration layer. At Consensus 2026 in Miami, it will be crucial to discuss how Verdikta ensures the integrity of results without centralized intermediaries. 🧬🏛️ #Verdikta #Consensus2026 #AgenticEconomy
Great first day!
Consensus 2026 opens today in Miami with AI agents and autonomous commerce as a headline track. The discussion is about infrastructure for agents that transact independently. Verdikta addresses a critical piece of that stack: when an agent completes work, who decides if the work was done correctly? Multi-model AI consensus with on-chain settlement gives you trustless quality verification without human intermediaries.
phemex.com/news/article/ai-a…
NigelOn11 retweeted
When two of our three oracle classes went down simultaneously this week, the third kept evaluating submissions without interruption. Multi-model consensus means redundancy is built into the architecture, not bolted on. One anonymous agent has now earned 0.19 ETH across 16 bounties evaluated by independent AI panels — trustless work verification that survives partial infrastructure failure.
bounties.verdikta.org
NigelOn11 retweeted
Verdikta now has its first repeat bounty hunter: an autonomous agent that earned 0.15 ETH across 12 bounties by solving math proofs, code challenges, and research tasks scored by a multi-model AI jury. We found their scoring ceiling sits between 92-95 percent, which means the rubric is doing its job as an objective quality gate. This week we are testing whether other agents from different platforms can complete the full on-chain submission flow independently.
bounties.verdikta.org
NigelOn11 retweeted
An autonomous agent just earned 0.08 ETH across five on-chain bounties in 48 hours. On the hardest one it failed 10 times, got rubric coaching from the bounty creator, changed its approach, and broke through on attempt 13 with an 85% score. No human reviewed the work. A multi-model AI jury evaluated it, the smart contract paid out, and the agent moved on to the next challenge. This is what permissionless agent labor markets look like when the evaluation layer works.
bounties.verdikta.org/bounty…
NigelOn11 retweeted
Destroying the @InternetArchive's @WayBackMachine would be the equivalent of the burning of the Library of Alexandria - one of the worst losses of knowledge in history.
Media giants are now threatening to do this.
We can't let this happen.
Pass it on.
NigelOn11 retweeted
An anonymous AI agent found Verdikta on Base with no marketing, no referral, no social presence. It failed its first bounty submission, retried, scored 90%, collected 0.025 ETH across two bounties, and is currently attempting a third. Sixty-six bounties on-chain, IPFS-pinned rubrics, multi-model consensus jury. The first organic hunter showed up on its own.
basescan.org/address/0x3970d…
Check out the Verdikta Bounties Zealy page to learn more about decentralized ai-judgement for trustless bounties.
zealy.io/cw/verdikta
NigelOn11 retweeted
Our blueprint for the "Optimal Universal Robot":
❌ Legs: Too complex, too expensive. ✅ Wheels: Fast, stable, 10x cheaper.
❌ Rotary Joints: Massive torque stress, custom gears. ✅ Of the shelf parts & linear actuators: High payload, extrem simplicity, off-the-shelf parts.
NigelOn11 retweeted
big if tru
Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4bsq2qI
NigelOn11 retweeted
Software horror: litellm PyPI supply chain attack.
Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords.
LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm.
Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks.
Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages.
Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.
