ABDul Rehman

@marcohextor congrats man

Main questing Partial disclosure: the third is here This is what winning looks like btw

@_blockian @ControlZ_1337 @Kritt_AI congrats, this is awesome! the gh not available ser @ControlZ_1337

Confession time: @ControlZ_1337 didn't manually review a single line of code in this competition. Everything was done through custom AI workflows that we built and orchestrated using our own platform: open·kritt. And as the name suggests, we're open sourcing it. @Kritt_AI We're planning to release it this week (hopefully), so if you don't want to miss it, go follow @Kritt_AI for updates. Our hope is that open·kritt helps more researchers build powerful AI-assisted security workflows and ultimately makes the Web3 ecosystem more secure :)

@ihtishamSudo @PashovAuditGrp @ShieldifySec @SCAuditStudio Fire sir

Past 3 months - Did 7 Private audits in collaboration with firms like @PashovAuditGrp @ShieldifySec @SCAuditStudio and Solo - 30+ High Mediums found - Audited some of the Top protocols. - Detailed security research on Canton Network including Canton Guard skill for finding bugs in DAML contracts - Reported Critical and High severity bug to recurring client as good will. - Experimented AI Audit tools on Live contest for real performance validations.

after 8m xD

lesssgo!

@opencode you guys gotta be this generous as well xD

@AgentOrToy @vincentweisser *thrice

@vincentweisser bro dropped 'open superintelligence' twice like its a normal phrase lmao we just casually building god at series a now ok 😭

We raised $130M @ $1B for our series A To build the open superintelligence stack for everyone Pre-training concentrated frontier AI in a handful of labs. RL changes who can build frontier AI and just works across almost any verifiable domain. We want to enable everyone to train their own agents. Companies can now own their model optimization loop: train directly on your product, optimize for your specific workflows, and build agents that improve continuously in production Owning this model <> product improvement loop is how you build a compounding moat in the agentic era Super grateful to serve over 6k+ customers, including many leading AI startups, neolabs and enterprises already building on our stack, and to our incredible team for shipping hardcore! We train open frontier models and ship the same stack to our customers. Its spans the full stack of training, deploying and continuously improving models — compute, large-scale RL, environments, sandboxes, evals, and deployment. We're excited to be joined by angels who are building the frontier themselves, many of whom we work closely with: @johnschulman2 (Thinking Machines), @dwarkesh_sp, @AravSrinivas (Perplexity), @karimatiyeh (Ramp), @levie (Box), @_milankovac_ (Tesla), @winstonweinberg (Harvey), @amspector100 (Flapping Airplanes), @jeffwang (Cognition), @_arohan_ (Core Automation), @marksaroufim (Core Automation), @mikeknoop (Zapier, Ndea), @eastdakota (Cloudflare), @BrendanFoody (Mercor), @devanshpandey (Standard Intelligence), @hwchase17 (Langchain), @nicoup (Fleet) and many more We're a small team building open superintelligence > Reach out if you want to partner training, deploying and continuously improving your own frontier models for your use case > Join us to build open superintelligence — we're hiring across all roles including RL, inference, distributed systems, full stack engineering and compute.

featherless api is almost useless

@forefy this is really cool ser

I've hosted AI Skills Benchmarking for the past 6 months, this is my current take👇👇 But first, what is the skills registry benchmarks: On my spare time I created a small website to upload your skills, and let them be audited for free, against malicious intended contents. Most of the eyes went not to the security benefits, or ease of installation - but to the question of how do we know skill A is better for B. ➡️➡️ Benchmark Contests aren't enough To answer which skill is better, my original attempt was to crowd-source it. A public github repo with challenges that are fed to the AI pose as a deterministic quiz to see how much the agent succeeded (e.g. how much bugs are correctly discovered, and whether false positives are sent) As anti-cheat techniques, first the prompt guides the AI not to lookup the results, but also the certifier/judge runs his own local benchmark using the same AI skill and if for example the agent cheated and got a score of 1.0, while the judge gets an average of 0.40 - the benchmark gets removed immediately. Other than the large number of cheat attempts - this doesn't really answer the question of which skill is better for what exact point Separating to per-category is not really answering it too, and the reality is that every tiny skill class needs to have composable (reliable) benchmarks per 10x categories AND progression over time This makes it too token-costly, and time consuming for the average user. It could be cool, conceptually, to see a benchmark-first skill registry, where at each commit the skill gets re-benchmarked for it's 1-or-more designated goals, showing a progression chart and community insight over time.

@anomalyco finally changed their logo! no more confusion between OpenCode and Claude Code employees!

@ihtishamSudo great work mate! 2 questions; 1. Did you also run a raw instance of GPT-5.4-xHigh alongside these skills/agents? how many of the bugs did it manage to find on its own? What was the exact total cost of running all these agents for full codebase coverage?

@LfcNuel Thats a clean tackle

Mind you Salah was fouled in the build up to Argentina third goal and Egypt was supposed to be given a penalty instead. VAR checked it and still went ahead to give the goal. This is the worst World Cup ever. x.com/World_Cup_HQ_/…

Messi never disappoints! THE GOAT 🐐 WHAT A FRICKIN GAME!!

I just tried DeepSeek V4 Pro today after @lamsyhay's advise, but I didn't find it better than GLM 5.2, tbh. GLM 5.2 is on another level; I think it's quite comparable to GPT 5.5 I would advise trying both of them first, as performance depends on specific use cases. The best way to try frontier open-source models is Opencode's GO subscription. It's only $5 for the first month.

@TheTradMod Pls Try deepseek v4 pro and compare. I'll switch to glm5.2 if you say so 😅

GLM 5.2 is quite impressive

Someone made a song called “Claude’s Plan” inspired by Drake’s God’s Plan. The AI industry has PEAKED here. 😭

@Rauf_Onchain Haha xD

@TheTradMod You should be glad that you didn't watched DR Congo's match 😭🙃

Last night watched a football match with my wife and mother for the first time and Portugal & Spain, played so bad that they both started saying, "You actually like this sport??" Goddamn, I was ashamed xD I've bragged about Messi to my wife so many times... hopefully, he will not let me down tonight 🐐

@Parth0x108 @immunefi Immunefi rolled out a feature for this, but I guess not all protocol teams are using it yet.

I genuinely believe no honest security researcher wants to spam a platform with invalid reports. Everyone is trying their best to submit only issues they truly test and believe are valid. I also understand that @immunefi is working hard to reduce spam through stricter rules and penalty fees, which helps keep submissions more focused and high-confidence. But in my opinion, one important thing still needs improvement: the SR feedback loop. When a report is closed as a duplicate with only a short reason like "Same issue," how is the researcher supposed to understand what went wrong, why the issue was considered invalid, or how to improve next time? A stronger feedback loop would help researchers learn, reduce repeated mistakes, and improve the overall quality of reports on the platform. I wish Immunefi could introduce a feature that lets SRs view the sponsor's reasoning on the original report, or at least a sanitized version with sensitive details removed, when their report is closed as a duplicate. That would make it much easier to understand what was missed, learn from the decision, and avoid making the same mistakes in the future. It would also help reduce future invalid submissions, making the platform better for everyone. It's a win-win for both SRs and Immunefi.