GPT-4 is the Dumbest Model You’ll Ever Use

Today’s top AI Highlights:

1. Sam Altman says ChatGPT is mildly embarrassing, GPT-4 is the dumbest model

2. OpenAI might be releasing a search engine+ChatGPT

3. Chat with data on your computer with images and voice commands

4. Convert any URL to an LLM-friendly input

5. Add code interpreting capabilities to your AI app

& so much more!

Read time: 3 mins

The rapid innovations in AI bring both excitement and concern. In a conversation with Stanford’s budding entrepreneurs, Sam Altman shared his insights on the future of AI and OpenAI’s approach to developing the technology responsibly. OpenAI’s team has repeatedly iterated the potential dangers of AGI, both catastrophic and subtle. Here’s what Sam Altman said about the current state and what the future of AI looks like:

1. AI is the Future: This is probably the best time to start a company. What you can do with AI is just going to get more remarkable every year and the greatest companies get created at times like this.

2. GPT-4 is the Dumbest: ChatGPT is not phenomenal, it is mildly embarrassing at best. GPT4 is the dumbest model any of you will ever have to use again.

3. Approach for Deployment: We believe in iterative deployment. If we go build AGI in a basement and then the world is blissfully walking and blindfolded, I don’t think that’s a good thing to do. It’s important to understand the way to put the product in people’s hands. Let the society co-evolve with the technology.

4. Does the Cost of AI Matter? Whether we burn 500 million a year or 5 billion or 50 billion a year, I don’t care. As long as we can create way more value for society than that and pay the bills, we’re making AGI. It’s gonna be expensive, it’s totally worth it.

5. Would AI recognize its flaws? The ability to recognize an error in reasoning and have some sort of introspection ability seems really important for me to pursue.

6. On GPT-5: We can say right now with a high degree of scientific certainty, GPT-5 is gonna be smarter than a lot smarter than GPT-4, GPT-6 is gonna be a lot smarter than GPT-5 and we are not near the top of this curve.

7. Building something smarter than humans: We are smarter than our previous generations not due to biological changes but via infrastructure improvements, like the internet and smartphones. This societal “scaffolding” will be further advanced by AI, allowing future generations to solve more complex problems and achieve more.

It seems like OpenAI is getting ready to launch its own search engine or some other AI product to search for information. Clues like the domain name “search.chatgpt.com” were found in their recent files, and there are whispers about a possible launch date of May 9th.

Key Highlights:

1. Domain Name in the Logs: The domain search.chatgpt.com was discovered within OpenAI’s SSL certificate logs, which are used to secure communication between websites and users. This strongly suggests that they are working on a search-related project.

2. ChatGPT Integration: The new search engine might use ChatGPT’s abilities to understand and respond to your searches in a more natural and conversational way.

3. Hints on Lex’s Podcast: In a recent podcast with Lex Friedman, Altman mentioned that creating a better version of Google isn’t interesting, but LLMs integrated with search can help people find, synthesize, and act on information. He hinted that ChatGPT could play a role in this.

4. What this means: This could give a tough competition to Google search, Bing and can even finish Perplexity AI and You.com.

Nvidia’s ChatRTX, the AI-powered chatbot app that lets you interact with your own data on your computer, using natural language, has received some cool updates. ChatRTX now supports new AI models, and lets you interact with images and voice commands, along with enhanced features for game modding through RTX Remix.

Key Highlights:

1. New AI Models: ChatRTX now supports Google’s Gemma and ChatGLM2 (a bilingual English and Chinese model). Till now, it was using Mistral 7B.

2. Image Understanding: Using OpenAI’s CLIP technology, ChatRTX can now analyze and understand images. You can give it images and ask questions about them, without needing complex metadata tagging.

3. Voice Interactions: Using OpenAI’s Whisper, ChatRTX can now understand voice commands in multiple languages. Simply speak your queries, and ChatRTX will respond with text-based answers.

😍 Enjoying so far, share it with your friends!

Share Unwind AI

1. Jina AI Reader: Converts webpages into a format easily understood by AI models. You can use it to feed information from any URL directly into your AI systems for tasks like research, summarization, or QA. The tool also automatically captions images on the webpage, allowing your AI to understand and interpret visual content.

2. Engine: Build a backend for your applications quickly and easily using natural language. It automatically generates the database schema and API code in either Typescript or Python, eliminating the need for manual coding or infrastructure setup. You can then deploy your backend instantly or connect it to your existing stack, giving you full control over your project.

3. AI Parquet: Interact with your local Parquet files using natural language, eliminating the need to write complex SQL queries. The tool processes your data securely within your browser, ensuring your information remains private and doesn’t leave your device.

4. Code Interpreter SDK: Integrate code interpreting into your AI applications. This means your AI app can now run and interpret code, enabling functionalities like automated data analysis, generative UI elements, AI tutoring, and even AI development assistance.

1. If people carry their phones everywhere, and AI applications on phones are excellent, I don’t understand why phones don’t win the race for “AI device” – not that there isn’t room for something like Meta glasses as a peripheral, but why would a dedicated AI device ever win? ~
   Ethan Mollick

2. HUGE problem for startups selling AI products to enterprises – buyers have no idea what is going on
   you think evals are a problem in research, but way more so in the real world: noisy, snake oil marketing, top-down pressure, lack of talent, focus on credibility as safe signal ~
   sarah guo // conviction

That’s all for today! See you tomorrow with more such AI-filled content.

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!

PS: I curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!

Share Unwind AI