- AI Fire
- Posts
- 🌐 Alibaba’s 'Hybrid' Models Also Thinking Budget
🌐 Alibaba’s 'Hybrid' Models Also Thinking Budget
Anthropic Economic Index on Devs
Wendy
April 28, 2025
Read time: 5 minutes
Imagine searching for products online but instead of hunting through endless ads and links, the perfect item simply finds you. And while you weren’t looking, tech giants like Google, Amazon, TikTok, and Instagram have been secretly battling behind the scenes.
What are on FIRE 🔥
IN PARTNERSHIP WITH BREWSTER CONSULTING GROUP
AI is everywhere - but are you using it effectively? Our AI Maturity Audit shows where you stand and builds your roadmap. If we can't find at least as much value as you pay us, it's free. No fluff - just real outcomes.
AI INSIGHTS
Anthropic analyzed 500,000 coding-related interactions across Claude.ai (the “default” Claude mode) & Claude Code (its new specialist coding “agent” ). The study focuses on automation vs. augmentation, AI usage by developer demographics, and types of coding tasks and languages.
Key Patterns Found in the Report:
Automation vs. Augmentation: Claude Code was more focused on automation than Claude.ai (79% vs. 49%), indicating that AI is increasingly taking over coding tasks with minimal human input.
Coders commonly use AI to build user-facing apps: While Python (14%) and SQL (6%) are used for backend and data-focused tasks.
The rise of "vibe coding": Developers increasingly just describe what they want in plain English and let the AI figure out how to build it.
Who’s using Claude to code: Startups are the main early adopters of Claude Code, while enterprises lag behind (33% compared to only 13% from enterprises)
How Software Use Differs from Other AI Use Cases:
AI use in software tasks is more automation-heavy.
A lot more Feedback Loop behavior (+18.3%)
Less Directive use (-11.2%), devs still want to review AI’s work before shipping.
The boundary between automation and augmentation becomes increasingly blurred with agentic tools. For example, the “Feedback Loop” pattern differs qualitatively from traditional automation, because it still requires user supervision and input.
Why It Matters: Coding is already one of AI’s strongest use cases - a possible early indicator of how other knowledge jobs could change soon. But we all know AI can’t replace all devs. So which software development roles will change the most, and which might disappear entirely?
PRESENTED BY IGNITION
This guide is your go-to resource for streamlining payments, improving cash flow, and keeping your business running smoothly.
What’s inside:
✔️ An actionable 8-step framework to create a seamless payment process
✔️ Expert strategies to reduce late payments and enhance your professional image
A well-structured payment system leads to smoother operations, happier clients, and long-term financial success.
TODAY IN AI
AI HIGHLIGHTS
🛒 ChatGPT is shaking up online shopping with non-ad-based search results, personalized product suggestions - over 1 billion searches in a week. Will this feature finally shift search from "you find it" to "it finds you"?
🙄 Interesting Fact: Google, Amazon, TikTok, Instagram, and Pinterest all fight to be shopping platforms, not just search or social.
Will this change how you shop online and impact Google's massive ad market? |
🎨 You can now generate images in Perplexity with a “cute and fun” user interface. The search engine just got support for Grok 3 and o4-mini, with o3 coming soon.
🚨 Researchers just secretly ran a massive, unauthorized AI experiment on Reddit users to change people’s minds on contentious topics. But it was conducted without Reddit’s consent.
🔍 Anthropic has mapped over 30M features in Claude 3 Sonnet, aiming to create a reliable “AI MRI” to diagnose models and better understand their “black box”.
💼 Microsoft’s new AI tools are sparking fears of job losses. As automation increases, workers in many industries could soon face being replaced by AI.
💰 AI Daily Fundraising: Goodfire has secured $50 million in Series A funding from Menlo Ventures and others, including Anthropic. The team aims to make AI models safer by ensuring they are understandable and controllable.
AI SOURCES FROM AI FIRE
NEW EMPOWERED AI TOOLS
🎥 Ztalk.ai is an app joining in video calls with real-time AI voice translation.
🕵️♂️ Competely AI Agent instantly find, analyze & track your competitors with AI.
📚 Shepherd turns multiple files into notecards, flashcards, smart notes,…
💻 RightNow AI 'V2.0' is a vibe coding platform for CUDA engineers.
🎓 notclass turns YouTube videos into your classroom with a little AI agent.
AI QUICK HITS
🤖 Sam Altman said ChatGPT’s personality is too sycophant-y & annoying…
🐛 OpenAI is fixing a ‘bug’ that allowed minors to create erotic conversations.
🚀 Microsoft's "Copilot for all" vision aims to democratize AI with custom agents.
📱 X’s social feed will get an algorithm update powered by xAI’s Grok model.
🧠 Duolingo will replace contract workers with AI, becoming an "AI-first" company.
AI CHART
Alibaba has just dropped a bombshell in the world of AI: Qwen3, according to early benchmarks, outperforms DeepSeek R1, OpenAI’s o3-mini & Google's Gemini 2.5 Pro in several key areas.
Most impressively, their largest model - boasting a colossal 235B parameters - even tops the charts in elite coding and reasoning tests. Yet, Qwen-3-235B-A22B, the biggest model, isn’t even publicly available yet. Intrigued?
Key Technical Features:
Switch between 2 distinct modes: reasoning and non-reasoning. You can control the "thinking budget".
Some versions use MoE (Mixture of experts) design for greater computational efficiency.
Trained on ~36 trillion tokens, supports 119 languages.
Outperforming Global Rivals:
Outshines o3-mini in advanced math benchmark and reasoning capability test.
Competitive with DeepSeek’s R1.
Surpasses o1 model on coding performance benchmark.
While these 6 smaller models are open to the public, the most powerful version - Qwen-3-235B - remains behind closed doors... for now.
DeepSeek R2 is expected to be a major leap forward after R1, which had already competed closely with Qwen2. Alibaba may possibly aim to launch it alongside DeepSeek R2 for maximum impact then!?
AI JOBS
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply