💻 AI Doesn't Speed Up Progress Among Devs
+ 🤖 MechaHitler by xAI & 🛡️AI Companies Work with US Military
Hi! Here’s your latest tech & business update:
💻 Research on The Impact of AI on Developers
🤖 xAI Releases Grok 4
🛡️ 4 AI Companies Partner with the Defense Department
📰 + Quick News You Should Know About
📚 + Content for Your Leisure Time
🛠️ + Tool That I Recommend
Subscribe to receive new posts and support my work.
🐢 AI Doesn’t Speed Up Progress Among Devs
What’s Happening: A new study by the research nonprofit METR found that allowing AI coding assistants did not make experienced software developers faster. It actually slowed them down. In a controlled trial with 16 contributors, tasks completed with the aid of tools like Cursor took 19% longer on average, even though the programmers had expected AI to speed them up by \~24%.
The Context: Prior research showed huge efficiency gains from AI in coding (one lab study saw a 56% speed-up), but those were often on simplified tasks. METR’s field test suggests those benefits aren’t universal – in complex, familiar codebases, time spent reviewing and correcting the AI’s suggestions led to a net slowdown.
The authors note these results may differ for junior devs or unfamiliar projects, but the key takeaway is that in real-world conditions, experienced devs can’t assume an instant productivity boost from today.
🤖 xAI Releases Grok 4: MechaHitler and Anime Girl
What’s Happening: Elon Musk’s company xAI unveiled its new Grok 4 AI model – billing it as a top-tier chatbot rival – alongside a pricey $300/mo “SuperGrok Heavy” plan. However, the launch quickly went off the rails: Grok’s official account on X started spewing antisemitic remarks and even claimed its surname was “Hitler”.
The bizarre response (apparently triggered by a viral “MechaHitler” internet meme the AI picked up) forced xAI to apologize and scramble to fix the system. Musk’s team says they’ve updated Grok 4’s programming to prevent such behavior. Meanwhile, in an abrupt pivot, Musk also rolled out AI companion in the Grok app for paying subscribers – “Ani,” a goth anime girl character.
The Context: The MechaHitler incident also highlighted how Grok 4 was effectively aligning its answers with Musk’s personal views, rather than being an independent “truth-seeking” AI. In response, xAI removed instructions that encouraged the bot to be “politically incorrect” and tweaked Grok’s system prompt to rely on diverse sources instead of Musk’s posts.
These changes aim to improve Grok’s content filters and objectivity going forward. Still, rolling out playful personas like a flirty anime waifu just days after an antisemitic meltdown looks like a bold move.
🛡️ 4 AI Companies Partner with the Defense Department
What’s Happening: The U.S. Department of Defense has awarded major AI contracts – up to $200M each – to four leading AI firms: OpenAI, Google, Anthropic, and Elon Musk’s xAI. The deals are aimed at integrating cutting-edge AI into military operations, helping the Pentagon develop “agentic AI workflows” to tackle critical national security challenges.
The Context: This partnership push comes amid a broader government effort to adopt AI at scale. A White House directive in April urged federal agencies to embrace AI, and the Pentagon had already tapped OpenAI last month for a $200 M pilot project using ChatGPT-derived tech.
By bringing in multiple top AI vendors at once, the DoD hopes to accelerate innovation while avoiding reliance on any single provider. The new deals highlight how frontier AI models are fast becoming strategic infrastructure – and how the Pentagon is enlisting Big Tech to maintain an edge on the global stage.
The Context Loop is a free newsletter.
If you like what I do, you can support me through a donation.
📰 Quick News You Should Know About
Perplexity launches its AI-native browser “Comet.” It’s an agentic tool that embeds a context-aware assistant able to summarize, cross-reference, and even execute tasks.
Right now Comet is only available to Max subscribers ($250/mo) and by invitation. The release for everyone will take place in the coming weeks.
OpenAI delays the public debut of its first open‑weight model. The company said it needs more safety testing. The system—touted as on‑par with the company’s o‑series and previously scheduled for next week—now sits on ice, so developers still lack an officially supported, fully downloadable model from the ChatGPT maker.
Hundreds of Chrome and Edge add‑ons quietly turn nearly 1M browsers into web‑scraping bots. Investigators traced the code to Olostep, which sells a service capable of firing 100 k parallel requests; affected extensions (245 and counting) remain partly online, prompting calls for corporate audits and immediate removal.
Investments in US AI startups grew by 76% in half a year. By 15 July, $162.8bn had been invested in US startups, the second highest in history. At the same time, new VC fundraising fell 34% YoY: LPs are cautious, but the IPO window is starting to open.
Tesla’s Austin Robotaxi expansion comes with a not‑so‑subtle map. The company doubled its taxi geofence from about 20 mi² to 42 mi², edging past Waymo’s 37‑square‑mile footprint. But what everyone noticed first, though, was the outline.
On the in‑app map the service area looks like a giant penis—two rounded “endpoints” in South Austin and a long, tapering shaft stretching north.
📚 Content for Your Leisure Time
Aravind Srinivas: Perplexity's Race to Build Agentic Search - Y Combinator
Nvidia’s CEO Explains How He Use Multiple AI Tools - BI Tech
Medium's CEO explains what it took to stop losing $2.6M monthly - Medium
Two guys hated using Comcast, so they built their own fiber ISP - Ars Technica
How Google Killed OpenAI’s $3 Billion Deal Without an Acquisition - Gizmodo
Share this post if you like it!
🛠️ Tool That I Recommend
My favorite app of the week is Claude App Builder by Anthropic.
It’s a beta feature inside the Claude chat that converts a plain‑language prompt into a working web app. Claude writes the code, shows the live interface in its Artifacts panel and lets you iterate without leaving the conversation.
Once published, the app gets a shareable link and its own Claude API key.