OpenAI develops web search capabilities

OpenAI integrated live web search into ChatGPT on Thursday last week, marking a significant shift in the AI chatbot's capabilities.

 By the CogX R&I team

November 08, 2024

The integration brings web-powered smarts to ChatGPT's existing interface, complete with interactive stock graphs, news citations, and even restaurant maps with pins. It also represents OpenAI's latest move to compete directly with traditional search engines like Google.

Credit: Getty Images

Key features include:

• Real-time web access integrated directly into ChatGPT's interface, with automatic triggering based on query relevance

• Interactive elements including stock graphs, news citations, and location-based maps

• A dedicated sources sidebar allowing users to verify information through original references

This launch arrives amid intensifying competition in AI-powered search. Google recently expanded its AI overview feature to over 100 countries, while Meta is reportedly developing its own search solution. Unlike Google's ad-supported model, OpenAI's ChatGPT search will remain advertisement-free, though the company acknowledges free users will face some usage limits.

OpenAI's search lead Adam Fry emphasised their commitment to publisher partnerships, telling The Verge they are "working very, very closely with all of those partners to understand how to use that content really responsibly." The company has secured agreements with major publishers including Hearst, Condé Nast, and News Corp, though they maintain these partnerships won't automatically prioritise content in search results.

More from OpenAI:

• OpenAI Pauses GPT-5 Development: OpenAI CEO Sam Altman has revealed that the highly anticipated GPT-5 won't be released in 2024. Instead, the company will focus on developing GPT-o1, a model designed for more complex tasks like mathematics and scientific reasoning.

  
  

• OpenAI continues to attract top talent: Caitlin Kalinowski, formerly of Meta, joins to spearhead robotics and consumer hardware, bringing her AR expertise from projects like Meta’s Orion. Also joining is Gabor Cselle, the co-founder of Twitter competitor Pebble.

Now read the rest of the CogX Newsletter

Getting Machine Learning Projects from Idea to Execution

Eric Siegel, former Columbia University professor and CEO of Gooder AI, outlines practical strategies from his new book, ‘The AI Playbook: Mastering the Rare Art of Machine Learning Deployment’, to help organisations turn machine learning projects into real-world successes.

Guest contributor: Eric Siegel

Humanity’s latest, greatest invention is stalling right out of the gate. Machine learning projects have the potential to help us navigate our most significant risks — including wildfires, climate change, pandemics, and child abuse. It can boost sales, cut costs, prevent fraud, streamline manufacturing, and strengthen health care.

But ML initiatives routinely fail to deliver returns — or fail to deploy entirely. They stall before deploying, and at great cost. One of the major issues is that companies tend to focus more on the technology than how it should deploy. This is like being more excited about the development of a rocket than its launch.

In this article, I offer an antidote: a six-step practice for ushering machine learning projects from conception to deployment that I call bizML. This framework is an effort to establish an updated, industry-standard playbook for running successful ML projects that is pertinent and compelling to both business professionals and data professionals.

… want to keep reading? Check out the full Q&A here on the CogX Blog

Also in the news

AI-powered scent teleportation: Imagine sharing the aroma of a freshly baked apple pie with a friend miles away, or experiencing the scent of a rainforest without leaving home. Thanks to AI, this future isn't far off. AI startup Osmo claims to have successfully used AI to teleport scents, digitally. In a recent demo, they showcased the digitisation and rematerialisation of a plum's essence.

AI PDF analysis is here: Anthropic's Claude AI just got a major upgrade. It can now analyse PDF files, processing not only text but also images, charts, and graphs. Whilst extremely useful, this powerful new feature is currently only available through a paid professional subscription or the model’s API.

Runway’s advanced AI camera control: Runway just launched a major upgrade for their AI video editor: precise camera controls. This new feature lets AI video creators add advanced pan, track, and zoom effects to their projects. Now, they can move the "camera" horizontally, arc around subjects, and even do those dramatic zoom-throughs popular in AI videos.

Meta develops a general-purpose touch encoder: Meta AI has unveiled Sparsh, a groundbreaking general-purpose encoder for vision-based tactile sensing. The system leverages self-supervised learning to train on a massive dataset of unlabeled tactile images, allowing it to generalise across various sensors and tasks. This means robots equipped with Sparsh can better understand and interact with the physical world, even in complex and dynamic environments.

AI model creates playable Minecraft from a picture

Credits: DECART

Fresh out of stealth mode and with $21 million in funding, Israeli AI company Decart is pushing the boundaries of artificial intelligence with Oasis, a breakthrough AI model that doesn't just generate in-game images but learns to recreate them in real-time. Their first tech demo? A fully playable version of Minecraft, generated entirely by artificial intelligence.

Trained on Minecraft gameplay footage, the system processes keyboard and mouse inputs to generate a playable environment on the fly, complete with physics, game rules, and interactive graphics — all running at speeds that put traditional AI video generators to shame.

The project demonstrates AI's growing capability to create interactive gaming experiences, operating with "zero latency" on a single NVIDIA H100 GPU and generating each frame in approximately 0.04 seconds. Significantly faster than existing AI video generators like OpenAI's Sora or Runway Gen-3.

Key technological achievements include:

• Real-time physics and game rules simulation without pre-programming

• Dynamic environment generation based on visual input

• Processing of keyboard and mouse inputs for immediate gameplay response

• Integration of Vision Transformer technology with diffusion models

In case you missed it

Clone, a robotics company based in Poland, have just unveiled Torso. The first bimanual torso actuated with artificial muscles: