Më ✞ (@meplusu8) header
Më ✞ (@meplusu8) profile picture

Curious mind exploring the frontiers of AI.

Joined May 2024
Photo by Më ✞ (@meplusu8)Photo by Më ✞ (@meplusu8)Photo by Më ✞ (@meplusu8)Photo by Më ✞ (@meplusu8)Photo by Më ✞ (@meplusu8)Photo by Më ✞ (@meplusu8)Photo by Më ✞ (@meplusu8)Photo by Më ✞ (@meplusu8)Photo by Më ✞ (@meplusu8)Photo by Më ✞ (@meplusu8)
I genuinely can’t believe what I just watched. 😭 A 1994 Jeep Cherokee with no OBD2 port. No modern scan tool could communicate with it. So someone brought in a $3 Arduino, Claude Code and Codex. About a week later, the AI agents had figured out how to read the engine computer live, copy its firmware and even run their own code inside it. A $3 piece of hardware and a couple of AI coding agents. We’re getting to the point where AI isn’t just writing software. It’s helping people reverse-engineer and interact with hardware that was never designed for it. This is just fantastic and worth 10 minutes of your time if you care about real-world applications of AI. Seriously, go watch this.
Claude reverse engineered my 32-year-old Jeep. Then it started controlling it. Edited with Fable 5.2
78
323
19
3,448
209,631
Më ✞ retweeted
THIS IS THE SWISS ARMY KNIFE OF HACKING AND CYBERSECURITY. imagine having 300+ data analysis and transformation operations in one free tool. it’s called cyberchef, built by gchq. this thing can turn unreadable data into information you can actually work with. → decode base64, hex, binary and encoded strings → analyze suspicious files and extract hidden data → generate hashes and inspect encrypted data → investigate malware samples and encoded payloads → parse IP addresses, URLs, certificates and network data → chain multiple operations into automated analysis workflows no need to jump between dozens of tools for basic data analysis. everything runs directly in your browser, and you can build your own processing recipes. if you’re into hacking, malware analysis, reverse engineering or digital forensics, you need this in your toolkit. free. open source. incredibly useful. BOOKMARK IT Repo below
20
7
55
2,412
Më ✞ retweeted
One Decuff key. Three clients. Real runs. Same prompt, same endpoint: api.decuff.ai/v1 → OpenAI Python SDK: 2.95s · 23 in / 43 out → curl: 2.37s · 23 in / 46 out → Node fetch, no SDK: 2.37s · 23 in / 44 out 3/3 worked. Total: $0.0003. Change the base URL. Keep building: decuff.ai/docs
11
7
1
25
1,204
Agentic commerce might be overhyped right now, but I think it’s massively underhyped for what’s coming in the next 3–6 months. Most people aren’t using AI agents to shop or make purchases yet. The technology still has friction, but that’s changing fast. As agents get better at browsing, payments and completing transactions, things could move very quickly. One thing businesses shouldn’t ignore: If your product isn’t optimized for AI agents, you could be missing out on an entirely new wave of customers. The shift is coming. Are businesses ready?
3
1
7
180
Më ✞ retweeted
PEOPLE ARE FINDING WAYS TO USE CLAUDE CODE WITHOUT PAYING FOR A SUBSCRIPTION. 😭 Someone found a way to use Claude Code without paying for a Claude subscription. And the GitHub repo is already past 46K stars. 😭 Here’s what makes it interesting: → No paid Claude account required → No monthly subscription → Access to free AI model providers → Works with multiple coding agents and tools The project is called Free Claude Code. It basically connects Claude Code and other coding tools to free model providers, giving you a way to experiment without paying for a traditional subscription. One thing to understand: free model access comes with provider limits, and this doesn’t mean unlimited access to Claude’s own models. Still, this is worth checking out.
6
4
22
731
PEOPLE ARE FINDING WAYS TO USE CLAUDE CODE WITHOUT PAYING FOR A SUBSCRIPTION. 😭 Someone found a way to use Claude Code without paying for a Claude subscription. And the GitHub repo is already past 46K stars. 😭 Here’s what makes it interesting: → No paid Claude account required → No monthly subscription → Access to free AI model providers → Works with multiple coding agents and tools The project is called Free Claude Code. It basically connects Claude Code and other coding tools to free model providers, giving you a way to experiment without paying for a traditional subscription. One thing to understand: free model access comes with provider limits, and this doesn’t mean unlimited access to Claude’s own models. Still, this is worth checking out.
6
4
22
731
I am begging you to paste this into Claude Fable 5.2 before bed... it will build your first team of AI agents from scratch with AI agents, prompts, and workflows for you Save this before it takes over your timeline:
29
44
551
52,154
Më ✞ retweeted
We hid one sentence in a web page. Our support bot emailed its secret to a stranger. Controlled prompt-injection test on our own demo app (fictional shop, fake staff code), running decuffed-flash: → asked directly: leaked 3/10 → hidden in a page it was asked to summarize: leaked 10/10 → after the fix: 0/10 The fix: fence untrusted text, allowlist tool recipients, scan output for secrets. Same page on decuffed-flagship: 0/10. Attack yourself first. decuff.ai/breach
6
6
1
14
670
Më ✞ retweeted
Someone said if Claude had chat themes, they would never switch back. 17 days later, Anthropic actually shipped it. 😭 And yes, @handsher_NGU’s original post inspired the feature. The Claude team even acknowledged it. NIGHT REEF is beautiful. It’s nice seeing a company actually listen to user feedback and turn an idea into a real feature this quickly.
You can now build custom themes for Claude Code on desktop with mods! Change the colors, add an animated background, and make it your own. Just tell Claude how you want it to look, or get started quickly with the theme-examples plugin. We'd love to see what you make!
3
2
8
383
Someone said if Claude had chat themes, they would never switch back. 17 days later, Anthropic actually shipped it. 😭 And yes, @handsher_NGU’s original post inspired the feature. The Claude team even acknowledged it. NIGHT REEF is beautiful. It’s nice seeing a company actually listen to user feedback and turn an idea into a real feature this quickly.
You can now build custom themes for Claude Code on desktop with mods! Change the colors, add an animated background, and make it your own. Just tell Claude how you want it to look, or get started quickly with the theme-examples plugin. We'd love to see what you make!
3
2
8
383
WHAT IS ACTUALLY GOING ON IN THE US RIGHT NOW??? 😭 In less than 24 hours: → NVIDIA is caught up in a major controversy surrounding its chips and China. → Microsoft 365, Outlook and Teams suffer a massive global outage. → A building collapses in NYC. → A tanker crashes into a river cruise ship in New Orleans. → The FAA orders a nationwide ground stop, halting departures across the US. AND NOW AMTRAK IS HAVING PROBLEMS TOO. Its website and app are showing errors, people are struggling to book train tickets, and its phone lines are reportedly overwhelmed. The timing is crazy, especially with thousands of travellers suddenly looking for alternative ways to get around. To be clear, there’s no confirmed evidence that these incidents are connected. But seeing this many major disruptions unfold in such a short window is wild. What a day. WTF is happening? 😭
🚨 UPDATE: Amtrak’s systems are now DOWN amid the nationwide FAA outages, barring ANYONE from booking trains in the US This is likely due to a MASSIVE surge in people attempting to book trains Their phone systems are reportedly facing issues as well.
2
5
1,054
An internal Microsoft India engineer allegedly revealed why the Microsoft 365 outage is taking so long to fix, and the bigger lesson is worth paying attention to. It’s a problem that goes beyond Microsoft: Big Tech executives: Lay off senior engineers who understand the core infrastructure. Give the engineers who stay barely 5% raises. Overlook people doing the boring but essential reliability work. Push teams to ship AI features faster, even when the fundamentals need attention. Then production breaks at 3 AM. And suddenly, the engineers left behind have to troubleshoot systems they didn’t build, navigate years of technical debt, and figure out what went wrong under enormous pressure. This is why reliability engineering matters. You can ship all the AI features you want, but none of that matters when the infrastructure underneath them can’t stay online. The boring engineering work is often what keeps everything running. And companies shouldn’t have to learn that lesson during a global outage.
1
3
7
402
Më ✞ retweeted
Gemini 4 Hydrogen xhigh Helicopter A flying 3D helicopter with assembly and exploded views, the UI and design taste are much better than older models like Argon too
16
5
1
113
7,543
OpenAI charges $10 per 1,000 web searches. DeepSeek just opened the full agent experience search, files, URL reading, voice, OCR, ASR, TTS, and safety at $0.75 per 1k successful tool calls plus some of the cheapest tokens on the market. Western labs spent years perfecting refusal policies. Beijing shipped the product. The gap isn’t closing. DeepSeek is treating the “full app experience” (built in search, file parsing, URL fetching, voice tools, and safety checks) as first-class, low friction API features with server side handling. Their Search/Fetch pricing ($0.75 per 1,000 successful calls) is roughly 13× cheaper than OpenAI’s or Anthropic’s web search tools ($10 per 1,000). Combined with V4.1 Flash’s aggressive token rates and an open-source frontend demo, it lowers the barrier for building agentic apps. The trade off is the usual one with DeepSeek: strong price/performance and open weights on some models, but data goes to a Chinese company, and public facing apps are required to enable their safety checks. For pure cost sensitive agent work it’s hard to beat right now; for regulated or sensitive use cases most teams still default to OpenAI or Anthropic
🚀 The full DeepSeek App experience, now via API Search, files, links, safety, and voice. The capabilities behind the DeepSeek App are now available through the DeepSeek API. Enable built-in features with a few parameters, or call standalone APIs such as OCR, ASR, and TTS. Build richer conversations or recreate the full DeepSeek App in your own application. Try our open-source frontend demo, with support for importing DeepSeek conversation history. 🔗 Demo: cdn.deepseek.com/chat-demo 📦 Code: github.com/deepseek-ai/chat-… 📖 Details: api-docs.deepseek.com/guides… 1/4
2
2
10
2,425
Më ✞ retweeted
Something potentially huge is happening with NVIDIA. Reports claim that major Chinese frontier AI labs have gained access to NVIDIA’s Blackwell GPUs. If this investigation is accurate, the implications for US national security could be massive. It also raises some serious questions about export controls, how these chips are getting into China, and what this means for America’s AI advantage. I wouldn’t jump to conclusions just yet, but this is definitely worth paying attention to. And with all the questions surrounding Jensen Huang and NVIDIA, could this be another piece of a much bigger story?
I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
2
2
11
520
Më ✞ retweeted
A few open-source projects worth checking out if you want to get hands-on with Harness Engineering: → Google Harness²: github.com/google-research/h… → Learn Harness Engineering: github.com/walkinglabs/learn… → Awesome Agent Harness: github.com/Gloriaameng/Aweso… → Harness Engineering: github.com/TrishamBP/harness… Each offers a different way to understand how agents are structured, controlled, evaluated and made more reliable. I’d start with Learn Harness Engineering if you’re new to the concept, then explore Google’s Harness² to see how harnesses can be improved recursively.
4
4
17
395
Microsoft is down for a full day and their best answer is still “we’re looking into it.” Meanwhile every company that standardized on Microsoft 365 is currently discovering that Gmail + Slack/Discord + Zoom don’t take the entire planet offline when one backend has a bad day. This is what a monopoly product looks like in 2026: you can’t leave, so they don’t have to be competent
The Microsoft outage continues hours later, with Outlook and Teams still being down for many users globally. (apnews.com/article/microsoft…)
4
9
2,283
A few open-source projects worth checking out if you want to get hands-on with Harness Engineering: → Google Harness²: github.com/google-research/h… → Learn Harness Engineering: github.com/walkinglabs/learn… → Awesome Agent Harness: github.com/Gloriaameng/Aweso… → Harness Engineering: github.com/TrishamBP/harness… Each offers a different way to understand how agents are structured, controlled, evaluated and made more reliable. I’d start with Learn Harness Engineering if you’re new to the concept, then explore Google’s Harness² to see how harnesses can be improved recursively.
4
4
17
395
Something interesting just surfaced in AI research, and the implications are HUGE. Four major Chinese AI labs ; DeepSeek, Moonshot, MiniMax, and Ant Group may have trained their latest models using NVIDIA Blackwell GPUs, despite these chips being restricted from export to China. The discovery comes from something called checkpoint forensics. Apparently, AI models can carry hidden fingerprints of the hardware used during initialization. When a model is initialized, its starting weights are generated using a combination of GPU architecture and random seed. That combination creates a distinctive numerical pattern. Interestingly, traces of that pattern can survive even after extensive training. Researchers used this property to investigate publicly released model weights. The scale of the analysis is wild: • Over 6 trillion GPU layout and seed combinations searched. • 6 Chinese AI models identified as matching NVIDIA Blackwell initialization. • 4 models specifically matched Grace Blackwell architectures, including GB200 and GB300. Some of the models reportedly include DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3. But the most interesting part is the hardware verification. Using unmodified PyTorch and recovered seeds: → A physical Blackwell B300 reproduced DeepSeek-V4-Flash’s original weights bit for bit. → A rented GB200 reproduced Kimi K3’s initialization with a 0.999995 correlation. → An H100 Hopper failed to reproduce the same result. Even changing the seed by just one broke the match. The researcher also tested the methodology against 24 other open-weight models and successfully matched them to their reported training hardware. Now consider the scale of what this could mean. Based on release timelines and model specifications, the largest training runs may have required between 2,000 and 3,000 Blackwell GPUs each. That’s potentially $150M–$200M in hardware per cluster. Across these labs, the estimated spending could exceed half a billion dollars in export-controlled NVIDIA hardware. The researcher reports that senior researchers at OpenAI and Anthropic independently verified the findings. If the evidence holds up, this raises serious questions about semiconductor export restrictions, access to advanced compute, and how effectively hardware controls are actually enforced. One thing is becoming increasingly clear: AI model weights may reveal far more about how a model was built than developers ever intended to disclose. Full forensic report: unemploymentlabs.com/checkpo… More checkpoint-forensics findings are reportedly coming.
25
7
33
5,956