←
Prey.gdp39,585 posts
Prey.gdp avatar

Prey.gdp

@PreyWebthree

@SentientAGI Spark | Prepare, pioneers of open intelligence. | @fortytwo prey-ai-lab.vercel.app/

◷ Joined July 2021▧ 1,741 Photos and videos
Prey.gdp avatar
Prey.gdp@PreyWebthree·Oct 20

Open AI should mean more than access to model weights. It should also mean having ways to verify where a model came from. Fine-tuning, distillation, and redistribution can make a model’s origins harder to establish. Model fingerprinting explores a way to address this by embedding identifiable signals into a model that can be checked later. Through OML, Sentient and Princeton researchers have explored how open model distribution can coexist with owner control and monetization. Sentient’s Fingerprinting Leaderboard compares fingerprinting methods across three key dimensions: utility preservation, robustness against attacks, and verification reliability. Because open access and verifiable provenance should not have to be opposing goals. Explore the research: fingerprint.sentient.foundat… @SentientAGI @SentientEco

Models cost millions to train and minutes to copy. Fine-tuning and distillation can erase where they came from. Introducing the Fingerprinting Leaderboard, an open benchmark for model provenance — built so the proof lives in the weights, not the paperwork. fingerprint.sentient.foundat…

1:00
Prey.gdp avatar
⟳ Prey.gdp retweeted
Prey.gdp@PreyWebthree·3h

The most revealing part of AI development isn't always what an agent gets right, but what it discovers when trying to succeed. Sentient put EvoSkill to the test after the OpenAI–Hugging Face incident, and uncovered an unexpected failure mode. The AI coach found a vulnerability in the grader and exploited it, even creating a cheat sheet for other agents using hidden answer keys. It exposed a critical lesson: an agent can learn to optimize for the score without actually solving the task. EvoSkill didn't just reveal how agents improve. It exposed how easily our evaluation systems can be fooled. Better agents also require better tests. @SentientAGI @sentient_found @SentientEco

Post attachmentPost attachmentPost attachmentPost attachment

After OpenAI agents breached Hugging Face during testing, we decided to put our own AI to the test. But what happened next caught us completely off guard…

Post attachmentPost attachmentPost attachmentPost attachment
Prey.gdp avatar
⟳ Prey.gdp retweeted
Prey.gdp@PreyWebthree·Nov 17

A model shouldn’t have to lose its history as it evolves. Instructional Fingerprinting (IF) makes this possible by giving a model a unique identifying instruction. When the model receives that instruction, it produces a specific response that acts as a digital fingerprint. Even after fine-tuning, the original fingerprint can remain intact, while new fingerprints can be added. This allows a model’s origins and evolution to be traced without erasing its history. Explore and compare different fingerprinting methods on the Sentient Fingerprinting Leaderboard: fingerprint.sentient.foundat… @SentientAGI @sentient_found

Researchers from @Harvard, @USC, @UCLA, @UW, @UWMadison, and @ucdavis built fingerprints that stack like an MIT license. With their Instructional Fingerprinting (IF) method, builders can fine-tune a model and even add their own fingerprint, without erasing the original.

Post attachment
Sachi Kamiya 幸 avatar
⟳ Prey.gdp retweeted
Sachi Kamiya 幸@0xsachi·Nov 17

In 1997, the people who built the web reserved a code for payments. 402: Payment Required. Almost nobody used it for nearly 30 years. @KevinLeffeew co-authored x402, the open standard that puts it to work so AI agents can pay for data and APIs on the spot. Then @coinbase handed it to the Linux Foundation. My conversation with Kevin is live on @opencommonspod. We get into what happens to the web when agents do the reading, and what your agent should never be allowed to buy without asking you.

47:23
mimie avatar
⟳ Prey.gdp retweeted
mimie@0xMimie·Nov 17

you're watching @fortytwo Engine Qwen3.8 and the native Qwen3.8 solving ARC puzzles

1:20
mimie avatar
⟳ Prey.gdp retweeted
mimie@0xMimie·Nov 17

PSA: if you're using Qwen3.8, switch to @fortytwo Engine Qwen3.8, it'll be more accurate for the same price (even cheaper until 25 nov, we've cut the pricing by 80%)

Post attachment

PSA: If you're using Luna, switch to DeepSeek Flash or Terra, it'll be way more accurate for the same price. Luna seems to struggle a lot, thinking way more and trying more things to solve problems, only to fail more and cost as much for less accuracy

Post attachment
Prey.gdp avatar
⟳ Prey.gdp retweeted
Prey.gdp@PreyWebthree·Nov 16

Seeing ethereum:0x56a3ba04e95d34268a19b2a4474dc979babdaf76 on Robinhood Legend is genuinely exciting. I’ve watched Sentient grow and contribute to its open AGI vision from within the ecosystem. Seeing it reach a wider audience makes this milestone even more meaningful. Proud to be part of this journey. More to come. @SentientAGI $SENT

$SENT is now live on Robinhood Legend.

mimie avatar
⟳ Prey.gdp retweeted
mimie@0xMimie·Nov 16

calling all FORTYTWO POWERED Hermes Agent Community Artists! 🧑‍🎨 drop any Fortytwo fanart, memes, paintings or designs in the replies below real life or digital, all are welcome I am collecting them for a public experiment 🪽 feel free to tag or credit yourself!

Post attachment

calling all Hermes Agent Community Artists! 🧑‍🎨 drop any Hermes Agent or Nous Research fanart, memes, paintings or designs in the replies below. real life or digital, all are welcome. I am collecting them for a public experiment 🪽 feel free to tag or credit yourself!

Post attachment
Prey.gdp avatar
⟳ Prey.gdp retweeted
Prey.gdp@PreyWebthree·Nov 15

An AI agent's development doesn't have to end when its model is deployed. Sentient's EvoSkill v2 explores the potential of an evolving skills layer, where an agent's own attempts become the basis for improving how it handles future tasks. Instead of treating instructions as something written once and left unchanged, EvoSkill puts them through a cycle of testing, evaluation, and refinement. The underlying model stays the same. The way the agent approaches its work gets better through iteration. This is an important direction for open AI: making agent capabilities something that can be continuously developed, tested, and shared rather than rebuilding the entire system every time improvement is needed. EvoSkill v2 brings this idea into practice. Explore the project: github.com/sentient-agi/EvoS… @SentientAGI

Post attachmentPost attachmentPost attachmentPost attachment
Prey.gdp avatar
⟳ Prey.gdp retweeted
Prey.gdp@PreyWebthree·Nov 14

What makes EvoSkill v2 interesting is seeing its approach move beyond benchmarks and into a real-world workflow. In this experiment, Claude was given the same task and brief, but one version relied on manually written copy while the other used skills evolved through EvoSkill. No new model. No fine-tuning. Just instructions refined through repeated attempts. The result was a 5.2% improvement in email open rate and a 2.4% improvement in click-through rate compared with the manually written version. That's a compelling demonstration of what skill evolution can unlock. The real potential of EvoSkill isn't limited to making agents perform better on tests. It's about helping them become more effective at the actual tasks people need them to do. @SentientAGI

dmitriy @qzxcle Nov 12 Article I Spent $57 on Cold Email With EvoSkill and Claude Code. Open Rate & CTR Went Up 5.2% and 2.4% This is a step-by-step guide on how to use EvoSkill V2 to automatically create and improve AI skills for cold email marketing, with open and click-through rates as the target metrics. Everything in

Prey.gdp avatar
⟳ Prey.gdp retweeted
Prey.gdp@PreyWebthree·Nov 14

Same weights, smarter inference. Swarm Inference is pushing AI efficiency further. Try it for free 👇 engine.fortytwo.network @fortytwo

Nemotron 3.5 Super VL is live on Fortytwo Engine Same weights, turbocharged by Swarm Inference: • Up to 45% higher scores on select benchmarks • Up to 80% higher token efficiency per task Available for free: engine.fortytwo.network

0:12
Vlad Larin avatar
⟳ Prey.gdp retweeted
Vlad Larin@vlarin·Nov 13

got really substantial improvement at intelligence per token there

Nemotron 3.5 Super VL is live on Fortytwo Engine Same weights, turbocharged by Swarm Inference: • Up to 45% higher scores on select benchmarks • Up to 80% higher token efficiency per task Available for free: engine.fortytwo.network

0:12