Meng Li

Codex .codex config deep dive. Latest version structure, key settings, and my personal setup. What’s your go-to Codex profile? #Codex #OpenAI #Claude open.substack.com/pub/aidisrupti…

Last month, Anthropic launched a collaborative initiative called Project Glasswing, aimed at hardening AI models before they become capable of attacking critical software. Over the past month, Anthropic and roughly 50 partners, using Claude Mythos Preview, have identified more than 10,000 high-severity or critical vulnerabilities across some of the world's most important software. The bottleneck in software security has shifted from the speed of finding vulnerabilities to the speed of manual verification, disclosure, and patching. The results have been validated across multiple fronts. Most partners each found hundreds of high-severity vulnerabilities within a month, with several reporting a more than tenfold improvement in discovery efficiency. Cloudflare identified 2,000 vulnerabilities in its critical systems, 400 of which were high-severity or critical, with a false positive rate lower than that of manual testing. The UK AI Security Institute called it the first model capable of fully solving both of its network attack-defense simulation environments. Mozilla fixed 271 vulnerabilities in Firefox — more than ten times the number fixed using Claude Opus 4.6. At one partner bank, the model also helped intercept a $1.5 million fraudulent wire transfer. In the open-source domain, Anthropic used the model to scan over 1,000 open-source projects, estimating that it uncovered 6,202 high-severity or critical vulnerabilities, with nearly 3,900 expected to materialize based on current true positive rates. In one case, a vulnerability was found in wolfSSL, a cryptography library used by billions of devices, which could have allowed an attacker to forge certificates and create highly convincing fake banking or email websites. However, Anthropic acknowledges that finding vulnerabilities has become trivial, while patching remains slow. Some open-source maintainers, overwhelmed by the volume, have even asked for a slower disclosure pace. Due to the lack of sufficiently robust safeguards against misuse of such models, Anthropic has not yet publicly released a Mythos‑level model. However, it has launched a public beta of Claude Security for enterprise customers, which has patched more than 2,100 vulnerabilities in three weeks. #Anthropic aidisruption.ai/p/two-mcps-cla…

ChatGPT developer OpenAI is working with investment banks to prepare for an IPO, with plans to confidentially file a prospectus with regulators in the coming days or weeks. Goldman Sachs and Morgan Stanley are among the firms assisting in the drafting of IPO documents, with a target to go public as early as September. However, the plans remain subject to change. OpenAI was valued at $852 billion in its most recent funding round. Earlier this week, OpenAI won a legal dispute with co-founder Elon Musk, removing a major obstacle to its public listing. During the trial, Musk had accused OpenAI of effectively “stealing charitable assets” by transitioning from a non-profit to a for-profit company. OpenAI argued that Musk had originally supported the transition and only broke with Sam Altman and others after realizing he would not have full control. The jury ultimately sided with OpenAI, and the judge dismissed Musk’s claims. Musk has said he will appeal. Still, OpenAI faces significant challenges, chief among them whether it can generate enough revenue to support its massive data center investments. Altman has been pushing for a swift IPO, but CFO Sarah Friar has reportedly told senior executives that more time may be needed. If all goes according to plan, alongside SpaceX’s upcoming IPO and Anthropic’s exploration of a large-scale public offering, this year could see the largest total IPO fundraising volume in history. #OpenAI #ElonMusk aidisruption.ai/p/anthropic-to…

This week, SpaceX formally submitted its S-1 prospectus to the U.S. Securities and Exchange Commission, preparing to list on Nasdaq under the ticker SPCX in what could become the largest IPO in history. Market rumors have previously valued the company at as much as $1.75 trillion, with a record-breaking fundraising target of $75 billion. The prospectus reveals SpaceX’s financials for the first time. In 2025, the company generated $18.67 billion in revenue, of which over $11 billion came from the Starlink satellite internet business. However, it posted a full-year loss of more than $4.9 billion, while capital expenditures surged from $11.2 billion in 2024 to $20.7 billion. xAI, which has since merged with SpaceX, contributed $3.2 billion in revenue last year but recorded an operating loss of $6.4 billion. Thanks to super-voting shares, Elon Musk will maintain 85% control over the company. SpaceX has told investors that its mission is to make humanity a multi-planetary species, and it views sending servers to space to build "orbital AI compute" as a massive opportunity. As is standard practice, the prospectus also lists risks, acknowledging that the markets it envisions for AI, orbital, lunar, and interplanetary transport are still nascent and may not develop as expected. It also admits that high debt levels could materially harm its financial condition. Additionally, as part of an AI compute partnership, Anthropic will pay SpaceX $1.25 billion per month through May 2029. Early yesterday morning, SpaceX successfully launched its 12th Starship test flight from its Starbase facility in Texas, completing the maiden flight of the all-new Starship V3. The launch had been aborted the previous day due to a hydraulic pin failure. The Starship V3 vehicle that flew consists of the upper-stage Ship 39 and the Super Heavy booster Booster 19. Standing taller than a 40-story building, it is the largest and most powerful rocket ever flown. #SpaceX #ElonMusk aidisruption.ai/p/googles-sear…

Making PPTs was never hard—just inefficient. Gemini + Google ecosystem fixed that. Which of these 5 workflows fits your pain point best? #PPT #Google #Gemini open.substack.com/pub/aiquill/p/…

The new intern spent a week lost in 100k lines of code. Sound familiar? This tool turns any codebase into an interactive graph so you can stop digging and start exploring. What would you map first? #graph #AI open.substack.com/pub/pythonlibr…

UEC has launched production at Burke Hollow after a 14-year cycle and is scaling multiple ISR projects in Texas and Wyoming with permitted plants, zero debt, and $818M in cash as US policy boosts domestic uranium supply. Is this the start of a multi-year production ramp for UEC? $UEC $UUUU $SMR open.substack.com/pub/robopub/p/…

ServiceNow trades at $100–102, down 60%+ from its high, yet offers 80–120% upside as the enterprise AI governance layer—would you buy the fear? $NOW $PATH $SNOW open.substack.com/pub/frontierst…

Nvidia’s $81.6B quarter shows AI’s first wave went to GPUs and HBM. Now, analog chips are getting the second wave. Is your AI bet ready for the power and signal war beneath the compute layer? $NVDA #GPU #HBM open.substack.com/pub/chippub/p/…

Welcome to The Explainer! Today, we're diving deep into a comprehensive 360-degree investment audit on Uranium Energy Corp. (UEC), examining its business and market trend analysis. This detailed financial analysis helps evaluate if UEC stock is a genuinely viable, long-term investing opportunity among uranium stocks in the current stock market analysis. #uuuu #uranium #nuclear #cleanenergy #energy #defense #stockmarket $UEC $UUUU $SMR youtu.be/2vgSN3DZlYM

Welcome to The Explainer! Today, we're evaluating ServiceNow as a potential long term investing opportunity, examining its rock-solid business fundamentals against a recent price drop. This detailed analysis explores its strategic positioning within the enterprise technology sector. We also consider how artificial intelligence plays a role in its business growth to help guide your decisions in the current stock market. #saas #ai #agent #servicenow #enterprise $NOW $PATH $SNOW youtu.be/PZ5QX4LDkCI

Anthropic just turned profitable for the first time ever. OpenAI still loses $1.22 on every dollar. What went wrong? #Anthropic #OpenAI open.substack.com/pub/aidisrupti…

Microsoft is in talks to supply its Maia AI chip to Anthropic, though no deal is final. Maia isn’t yet available to Azure customers, so adoption by Anthropic would signal external validation and a step toward commercializing Microsoft’s self designed AI silicon, where it lags Amazon and Google. Anthropic needs compute badly for Claude and Claude Code, already using Nvidia, AWS Trainium, and Google TPU. Adding Maia would diversify further, but near term impact on Microsoft’s financials looks small. What could change Microsoft’s chip ambitions faster than an Anthropic deal? #Microsoft #Anthropic #Google open.substack.com/pub/aidisrupti…

OpenAI plans confidential IPO filing with SEC as soon as Friday targeting September listing. Same day, SpaceX filed new offering docs and Anthropic disclosed strong profit outlook. Anthropic’s Q1 operating profit expected this June; April enterprise paid client share hit 34.4% vs OpenAI’s 32.3%. Internally, Altman pushes September IPO while CFO Friar hesitates. Co founder Karpathy joined Anthropic yesterday. Which AI firm would you trust most as a long term public investment? #OpenAI aidisruption.ai/p/googles-sear…

DeepSeek funding has surged to 70B yuan. Management tells potential investors they prioritize breakthrough AI research over short term commercialization. Meanwhile DeepSeek Code is real. Senior researcher Deli Chen posted hiring for Harness team and Code Harness. Team lead may be Cui Tianyi, former TSY Capital co founder who joined DeepSeek in March. Job posts include Agent Harness PM and engineer. Requirements include deep use of Claude Code, Cursor, Manus. Deli Chen says build Code Harness from scratch. DeepSeek Code appears to be coming. What would you want this coding tool to do first? #DeepSeek aidisruption.ai/p/two-mcps-cla…

65 lines from Karpathy just fixed AI coding. Accuracy jumps from 65% to 94%. Want your AI to stop guessing and start thinking? #Karpathy #AI #CLAUDE #Prompts open.substack.com/pub/aiquill/p/…

Discovered TuriX-CUA on GitHub: an open-source AI agent that controls your desktop like a human. No API per app needed. 80%+ success on Mac. Have you tried AI desktop automation yet? #AI #desktop open.substack.com/pub/pythonlibr…

Archer holds $1.78B cash but burns over $600M yearly as it races for FAA certification, with its $6 stock trading near fair value. Do you see the dilution risk or the defense pivot as the bigger swing factor from here? $ACHR $JOBY $EVTL open.substack.com/pub/robopub/p/…

We called ARM perfectly. Opened yesterday, up 15% then 16% today, 30% total. Our model saw AI's CPU re-pricing coming. But why did Zoom jump after hours? $ARM $ZM $AMD open.substack.com/pub/frontierst…

AMD's first 2nm EPYC Venice CPU is now in mass production via TSMC. Can AMD outrun Nvidia, Intel, and Arm in the Agentic AI race? $AMD $TSM $NVDA open.substack.com/pub/chippub/p/…