- The Shift
- Posts
- GPT-5... not so good after all
GPT-5... not so good after all
Plus, AI predicts protein locations in your body, Automate support with Tidio in Gmail or Zendesk, and more!
Hello there! Ready to dive into another upgrading, Mind-boggling, and Value-filled Shift?
Today we have:
🤖 Sam Altman Talks GPT-5 Glitches, 4o Comeback & Chart “Crime”
📨Automating 30% of Support Work in Gmail, Zendesk, and Beyond
🧬 AI Can Predict Any Protein’s Location in a Human Cell
🔨Tools and Shifts you cannot miss
🤖 Sam Altman Talks GPT-5 Glitches, 4o Comeback & Chart “Crime”
OpenAI’s GPT-5 rollout hit turbulence last week, sparking user frustration over technical failures, lower perceived intelligence, and a viral chart mishap. In a Reddit AMA, Sam Altman addressed the issues, promised fixes, and hinted at bringing back the beloved GPT-4o for Plus subscribers.
The Shift:
1 - Launch Day Router Failure Tanked Early Performance - A real-time router meant to choose between speed and deeper reasoning models failed on rollout day. This made GPT-5 responses feel “dumber” compared to 4o. Altman confirmed fixes are live and said future updates will reveal which model is answering queries.
2 - GPT-4o May Return for Plus Users - Users flooded the AMA asking for GPT-4o’s return, citing its personality and emotional intelligence. Altman admitted the company underestimated attachment to 4o and is exploring ways to make it available again for paying subscribers. He also promised to double the Plus rate limits during the transition.
3 - The “Chart Crime” Goes Viral - During the livestream, a benchmark chart showed a lower score with a taller bar, quickly becoming a meme. Altman called it a “mega chart screwup” but didn’t address it further in the AMA.
The GPT-5 launch shows that technical power alone doesn’t win user trust; experience, reliability, and emotional connection to the model matter just as much. For OpenAI, balancing innovation with stability will be critical to keeping its most loyal users engaged.
📨Automating 30% of Support Work in Gmail, Zendesk, and Beyond
If your support inbox feels like an endless loop of “Where’s my order?” and “What’s your return policy?”, Tidio Copilot can take the wheel.
This Chrome extension plugs into your existing platforms, Gmail, Zendesk, Intercom, or Gorgias, and instantly suggests AI-powered replies based on your own knowledge base.
How to Use Tidio Copilot
Install the Chrome Extension – Add Tidio Copilot from the Chrome Web Store.
Connect Your Platform – Log in to your preferred support tool (Gmail, Zendesk, Intercom, or Gorgias).
Enable Knowledge Base Access – Let the AI pull context from your company’s FAQs, help docs, or internal notes.
Start Handling Conversations – As tickets or DMs come in, Copilot will suggest ready-to-send responses. You can edit or approve them instantly.
Refine Over Time – The AI continuously learns from your team’s approved responses, making suggestions sharper with every use.
With instant AI-assisted replies, your team spends less time on repetitive tickets and more time building customer relationships. And since it’s free to start, you can test it today without risk.
🧬 AI Can Predict Any Protein’s Location in a Human Cell
Researchers from MIT, Harvard, and the Broad Institute have built PUPS, an AI that can pinpoint the exact location of almost any protein in any human cell, even ones it’s never seen before. By mapping proteins at the single-cell level, it could speed up disease diagnosis, drug discovery, and fundamental biology research.
The Shift:
1 - Single-Cell Precision - Unlike older models that give averaged predictions, PUPS predicts exact protein placement in individual cells. This means researchers can see how a protein behaves in a cancer cell after treatment.
2 - Two Models, One Prediction - PUPS merges a protein language model that understands amino acid sequences with an image inpainting model that reads cell structure and stress states. Together, they produce a highlighted cell image marking where the protein is likely located.
3 - Proven Accuracy Over Existing Methods - In lab tests, PUPS outperformed baseline AI tools with lower prediction errors across all proteins tested. It can detect localization shifts caused by mutations absent from major datasets like the Human Protein Atlas.
PUPS transforms protein mapping from a slow, resource-heavy process into a scalable, AI-driven workflow. For medical research and biotech, that’s a direct path to faster discoveries, sharper diagnostics, and accelerated treatments.
🔨AI Tools for the Shift
😴 History Sleep – AI-generated boring history lectures for better sleep. Perfect for winding down without screens or distractions.
📊 VoxDeck – AI presentation maker crafting stunning, personalized slides. Save time creating decks with polished, on-brand designs.
⚙️ SiliconFlow – One platform for all AI inference needs. Scalable and cost-effective solutions for running AI models in production.
💬 AI Customer Service Chatbot by BlueTweak – Automate customer support inquiries with AI-powered chat. Boosts efficiency while maintaining a personal touch.
📱 Wazzap AI – AI WhatsApp automation for agencies and businesses. Streamline client communication, follow-ups, and lead management.
🚀Quick Shifts
⚖️Nvidia and AMD will share 15% of China AI-chip revenues with the U.S. government as a condition for resuming exports of Nvidia’s H20 and AMD’s MI308 chips.
🚀 NASA and Google are developing the Crew Medical Officer Digital Assistant, an AI tool for diagnosing and treating astronauts in deep space, aiming for high accuracy without Earth contact and future broader healthcare applications.
🛍️ Pinterest CEO Bill Ready says fully autonomous AI shopping is still years away, emphasizing Pinterest’s role as an AI-powered assistant that delivers personalized, style-aligned recommendations while expanding AI in search, ads, and personalization tools.
🎨 Microsoft launches Copilot 3D, an AI tool that instantly turns images into ready-to-use 3D models for seamless integration into games, animation, VR/AR, and other creative applications with a single click.
♟️ OpenAI’s o3 won Kaggle’s AI Chess Exhibition, sweeping every match 4-0 and defeating Grok 4 in the final, with Gemini 2.5 Pro finishing third.
Reply