08 Aug 2025

GPT-5 a Banger or Clanger? Gemini Trespassing...

🧠 OpenAI's GPT-5 a Mixed Bag

OpenAI finally launched its much-anticipated GPT-5 model, and the reception has been... interesting. While OpenAI is touting it as its “most capable model yet,” with “breakthrough” upgrades in writing, coding, and more, the internet has been less than impressed, with some calling it “mid with a longer memory.”

Here's the lowdown:

What's new: GPT-5 comes in three flavors: GPT-5, GPT-5 Pro, and GPT-5 Mini. The base model is now available to all 700 million ChatGPT users, including those on the free tier. It boasts a 256k-token context window, Gmail/Calendar integration, and personalized “smart routing.”
The good: OpenAI claims GPT-5 hallucinates less, is less deceptive, and is more honest about its capabilities. It also delivers state-of-the-art performance on various benchmarks. (In our tests it still promised to edit a video and then - psyche! - told us to go try a video editor.)
The bad: Critics, including Elon Musk, natch, have been quick to point out that GPT-5 still lacks continuous learning and is far from AGI. Musk even went so far as to say that his own Grok 4 Heavy was “smarter two weeks ago.”
The ugly: OpenAI’s launch was marred by a “mega chart screwup” in which a misleading chart was used to showcase the model’s reduced hallucinations. The company has since corrected the chart, but the incident has raised questions about OpenAI’s internal processes.

Why it matters: GPT-5 is a significant release, but it’s clear that the AI race is far from over. While OpenAI has made a powerful new tool available to the masses, competitors like Anthropic, Google, and xAI are hot on its heels. The next few months will be crucial in determining who will lead the charge towards AGI.

Read more at AI Secret and The Rundown AI.

🔬 Drug Discovery: A New Frontier

Alphabet’s secretive drug discovery arm, Isomorphic Labs, is gearing up for human trials of AI-designed drugs. This marks a significant step forward in leveraging AI for medical advancements.

AlphaFold's Impact: Isomorphic Labs was born from DeepMind’s AlphaFold, an AI system renowned for its ability to predict protein structures and interactions with high accuracy. This capability is crucial for accelerating drug design and making it more precise.
Addressing Rising Costs: The cost of discovering new medicines has been on a steady rise. AI, with its potential to unlock an era of target abundance, could reverse this trend and significantly boost research and development productivity.

Why it matters: The application of AI in drug discovery holds immense promise for revolutionizing medicine. By speeding up the process and improving precision, AI could lead to breakthroughs in treating diseases that are currently difficult or impossible to cure.

🌳 AI for Wildlife Conservation: Listening to Nature

Google DeepMind has open-sourced an upgraded version of Perch, an AI model designed to help scientists analyze vast amounts of wildlife audio. This innovation aims to make tracking endangered species in diverse environments more efficient and effective.

Enhanced Capabilities: The new Perch model can process a wider range of species and environments, from dense forests to vibrant coral reefs. It leverages twice the training data compared to its 2023 predecessor, allowing it to disentangle complex soundscapes over thousands or even millions of hours of audio.
Practical Applications: Perch can answer critical questions, from species counts to detecting newborn animals. It also includes open-source tools that combine vector search with active learning, enabling the detection of species even with limited training data.

Why it matters: This open-source release is a significant breakthrough for wildlife conservation. AI’s speed and precision can provide conservationists with a crucial advantage in protecting ecosystems and saving species before they face extinction. Examples include accelerating honeycreeper monitoring in Hawaii by 50x and finding elusive bird populations in Australia.

💻 New Coding Tools ... and Challenges

The world of AI coding is rapidly evolving, with new tools emerging and existing ones getting significant upgrades. However, this rapid advancement also brings new challenges.

Cursor embraces GPT-5: Following OpenAI’s GPT-5 release, Cursor, an AI coding platform, quickly integrated the new model. This move provides Cursor with a “second” supply line, reducing its reliance on Anthropic’s Claude, which has been a costly constraint due to premium pricing and API throttling. This integration will serve as a stress test for GPT-5’s real-world performance in coding.
GitHub’s “vibe coding” challenge: GitHub’s CEO is grappling with the rise of “vibe coding,” a new trend where developers rely heavily on AI tools for code generation. This trend, coupled with new rivals, is challenging GitHub’s dominance in AI development tools.
LangChain Labs’ Open SWE: LangChain Labs has released Open SWE, an open-source, cloud-based asynchronous coding tool. This tool aims to automate planning, coding, and review work, allowing developers to focus on other tasks.

Why it matters: The increasing sophistication of AI coding tools is transforming software development. While these tools promise increased efficiency and automation, they also raise questions about the future of human programmers and the potential for new types of errors or vulnerabilities.

⚖️ Automating the Law

Thomson Reuters has launched CoCounsel Legal, an AI legal research tool designed to assist lawyers with various tasks, from drafting documents to summarizing cases and conducting legal research.

Efficiency for Legal Professionals: This tool aims to streamline the often time-consuming and labor-intensive process of legal research, allowing legal professionals to focus on more complex analytical and strategic work.

Why it matters: The integration of AI into legal research signifies a growing trend of automation in professional services. While it promises increased efficiency and accuracy, it also raises questions about the evolving role of legal professionals and the need for ethical guidelines in AI-powered legal assistance.

💰 Funding News

Chai Discovery Raises $70M: OpenAI-backed Chai Discovery has raised $70 million at a $550 million valuation. The company aims to commercialize its new model, Chai-2, for pharmaceutical companies, focusing on accelerating drug discovery and development.
Delve Secures $32M Series A: Delve, an AI-native compliance platform, announced a $32 million Series A funding round at a $300 million valuation, led by Insight Partners. Delve’s platform helps startups and businesses achieve compliance quickly, saving hundreds of hours of manual work.
Clay Raises $100M: Clay, a platform that enriches sales leads from over 130 data sources and automates manual research, has raised $100 million. This funding will likely fuel further development of its AI-powered data enrichment and automation capabilities.

Why it matters: These significant funding rounds highlight continued investor confidence in the AI sector, particularly in applications that promise to revolutionize industries like pharmaceuticals, compliance, and sales. The substantial investments underscore the potential for AI to drive efficiency and innovation across various business functions.

📈 Industry Trends

AI in Weather Forecasting: The Hong Kong Observatory reports that AI models are now outperforming conventional methods for medium-range weather forecasting. This indicates a growing trend of AI’s practical application in critical scientific fields.
The Rise of “Vibe Coding”: GitHub’s CEO is addressing the emergence of “vibe coding,” where developers increasingly rely on AI tools for code generation. This trend raises questions about the evolution of programming practices and the balance between human creativity and AI assistance. Plus the results / output are often very beta.
AI’s Impact on Hiring: Hiring for new graduates in Big Tech is down 25%, and for startups, it’s down 11%. This suggests that AI’s increasing capabilities might be influencing hiring trends, with companies potentially seeking more specialized or experienced talent.

Also, Microsoft’s research suggests that AI will significantly impact various jobs, leading to shifts in the workforce and the need for new skills. This trend highlights the ongoing discussion about AI’s role in the future of work.
AI and App Interaction: Google is implementing changes that will allow its Gemini AI engine to interact with third-party apps, even if users have configured their devices to block such interactions. This move points towards a future where AI systems are more deeply integrated into our digital ecosystems.

Why it matters: These trends collectively paint a picture of AI’s pervasive influence across various sectors. From scientific predictions to workforce dynamics and software development, AI is reshaping how we work, live, and interact with technology.

Read more at AI Secret, The Rundown AI, and TLDR AI.

✨ Quick Hits ✨

Microsoft Integrates GPT-5: Microsoft has integrated the newly launched GPT-5 model across its entire Copilot suite, making it available to all users. This move significantly enhances Copilot’s capabilities and broadens the reach of OpenAI’s latest model.
Meta Acquires WaveForms: Meta has acquired a16z-backed audio AI startup WaveForms. This acquisition aims to advance Meta’s emotional intelligence capabilities, suggesting a focus on more nuanced and human-like AI interactions.
Google’s “Big Sleep” Bug Hunter: Google DeepMind’s AI-powered bug hunter, “Big Sleep,” has reported its first 20 vulnerabilities in popular open-source software like FFmpeg and ImageMagick. This collaboration between Google’s AI team and Project Zero demonstrates the growing role of AI in cybersecurity.
OpenAI Increases Employee Stock Compensation: OpenAI is significantly increasing its employee stock compensation to counter Meta’s aggressive talent poaching. This move underscores the intense competition for top AI talent in the industry.
Waymo’s “Road Trips”: Waymo has initiated “road trips” to Philadelphia and New York City. While not a commercial launch, these trips suggest an expansion of Waymo’s autonomous vehicle testing and potential future service areas.
Apple Loses More AI Talent to Meta: Ruoming Pang, the executive leading Apple’s foundation models team, is departing for Meta, reportedly for a package worth tens of millions of dollars annually. This further emphasizes the fierce competition for AI expertise.

🛠️ New Tools & Opportunities

Floot: A new no-code platform with a full tech stack designed for entrepreneurs and non-coders to easily build functional web applications. (Floot)
Unicorns Club released traction-based startup rankings, community awards, the Unicorn Index, and Sparks, offering new ways to track and recognize startup growth. (Unicorns Club)
Patio: A DIY community platform for borrowing, renting, learning, and trading tools, fostering a collaborative environment for makers. (Patio)
Haimeta: A multi-modal platform that leverages over 20 AI models to create images, videos, 3D assets, and interactive spaces, providing a comprehensive creative suite. (Haimeta)
Heardly: Described as the “Fast Way to read Best Book,” suggesting an AI-powered summarization or content digestion tool for books. (Heardly)
CopyOwl: The “First AI Research Agent” capable of deep research on any topic with a single click, aiming to automate and accelerate research processes. (CopyOwl)
Flot AI: An AI tool that writes, reads, and remembers across various applications and web pages, enhancing productivity and information management. (Flot AI)
Perplexity Comet Shortcuts: A new feature in Perplexity’s browser (oh you don't have access yet? sorry) that allows users to create automated workflows triggered by simple commands, enabling real-time product review fact-checking. (Perplexity Comet Early Access)
Warp Lightspeed Plan: Warp, an AI coding agent, offers a new Lightspeed plan with generous monthly limits and access to top AI models like Opus, Sonnet, and GPT 4.1, catering to developers’ coding needs. (Warp)
PUPS (Protein Understanding and Prediction System): An AI system developed by MIT, Harvard, and the Broad Institute that predicts the exact location of virtually any protein within a single human cell, accelerating disease research and drug discovery. (MIT News)
NuMarkdown: An 8B parameter, open-source reasoning model for Optical Character Recognition (OCR), designed for advanced text extraction and understanding. (NuMarkdown GitHub)
Midjourney HD Mode: Midjourney now offers an HD mode for video generations, providing 4x more detail and professional-quality footage. (Midjourney)
North: Cohere’s secure and customizable AI agent platform, offering a robust solution for building and deploying AI agents. (Cohere North)
Kitten TTS: A new open-source text-to-speech model that generates realistic speech with multiple voice options, running locally without a GPU. (Kitten TTS HuggingFace)
Overlap: A tool that automatically clips the best moments from long podcasts and formats them for social media with captions and vertical orientation. (Overlap)
LMArena Search Leaderboard: Released a new leaderboard showcasing the best models for web searching, with o3-search, claude-opus-4-search, and gemini-2.5-pro-grounding identified as top performers. (LMArena Leaderboard)
Orchids: A design-first “vibe-coding” tool that builds beautiful websites and apps from text descriptions without requiring any coding. (Orchids)
Google Guided Learning: A new tool that provides step-by-step problem-solving with questions and quizzes, offering a more interactive learning experience. (Google AI Blog)