OpenAI's newest o3 and o4-mini models excel at coding and math – but hallucinate more often
Hallucination rates spike to 33% for o3 and 48% for o4-mini
OpenAI rolls out GPT-4.5 to Pro users only, faces shortage of GPU horsepower
Plus-tier members will have to wait until next week for access
OpenAI is struggling with ChatGPT-5 delays and rising costs
The internet is running out of training data
OpenAI debuts "o1" AI models, promising PhD-level reasoning in science and math
The company is so confident about o1's abilities that it reset the version number back to 1
Google Gemini is not as good at analysis as its hype says, other AI models also struggle
The big picture: If there's one thing that generative AI is supposed to be good at, it's analyzing the written word. However, two studies suggest that this ability may have been overhyped. One study demonstrates that Gen AI struggles with understanding long-form books, while another shows that these models find answering questions about videos challenging. This is something companies should consider as they augment their workforce with Gen AI.
OpenAI launches faster and free GPT-4o model – new voice assistant speaks so naturally you will think it's hoaxed
The multimodal model sets a new bar for chatbots
Forward-looking: OpenAI just introduced GPT-4o (GPT-4 Omni or "O" for short). The model is no "smarter" than GPT-4 but still some remarkable innovations set it apart: the ability to process text, visual, and audio data simultaneously, almost no latency between asking and answering, and an unbelievably human-sounding voice.
OpenAI transcribed over a million hours of YouTube videos to train its LLMs, Google engaged in same practice
Meta reportedly considered equally shady tactics, too
GPT-4 loses its position as "best" LLM to Claude-3 in LMSYS benchmark
Anthropic's local-scale model Claude 3 Haiku grabs 7th with it "GPT-4-class" performance
OpenAI's GPT-5, their next-gen foundation model is coming soon
GPT-5 needs to solve the reliability issues experienced by GPT-4 customers
If you teach a chatbot how to read ASCII art, it will teach you how to make a bomb
In context: Most, if not all, large language models censor responses when users ask for things considered dangerous, unethical, or illegal. Good luck getting Bing to tell you how to cook your company's books or crystal meth. Developers block the chatbot from fulfilling these queries, but that hasn't stopped people from figuring out workarounds.
Researchers prove they can exploit chatbots to spread AI worms
Hackers could deploy the worms in plain text emails or hidden in images
In context: Big Tech continues to recklessly shovel billions of dollars into bringing AI assistants to consumers. Microsoft's Copilot, Google's Bard, Amazon's Alexa, and Meta's Chatbot already have generative AI engines. Apple is one of the few that seems to be taking its time upgrading Siri to an LLM and hopes to compete with an LLM that runs locally rather than in the cloud.
Knowledge workers who don't use AI will fall behind, says top executive
Roles aren't going away, but they are fundamentally changing
OpenAI to regulators: Training AI models without copyrighted material is "impossible"
Can companies use publicly available data to train LLMs?
ChatGPT achieves the pinnacle of human intelligence, laziness, and developers are baffled
Developers are aware of the issue and are trying to curb the behavior
Google proposes Project Ellmann, a chatbot that intimately knows you
Ellmann would need access to Google Photos, browsing history, and probably more
While others rush to be first, Apple has quietly spent 'millions of dollars a day' on multiple AI projects
Work on its Ajax GPT model already exceeds OpenAI's GPT-3.5 by 25 billion parameters
Scientists recreate Pink Floyd's "Another Brick in the Wall" using patient brainwaves and AI
The technology could open the door to more effective brain-machine interfaces
Healthcare chain is using a GPT-4-based AI to generate medical records
Saving 12 minutes over human-made medical notes, hallucinations aside.
The AI chip market landscape: Choose your battles carefully
There are three markets for AI chips: training, cloud inference and edge inference. All of them are already fairly crowded
Scientists have developed a GPT model that reads human thoughts
It cannot interpret thoughts verbatim but can put those ideas into an accurate description
Microsoft and Epic partnership will see OpenAI's GPT-4 used to discover trends in medical records
And draft letters to patients
UK government pledges £900 million to build "BritGPT" exascale supercomputer to advance AI technology
The bleeding edge: In case you haven't noticed, AI technology is all the rage with the kids lately. If it's not Dall-E or Stable Fusion; it's ChatGP, Bart, Bing Chat, or some other company's large language model. Not wanting to get left in the dust, the UK sank a sizable investment into building the next exascale supercomputer.
Microsoft Copilot enables next-gen AI for Word, Excel and other Office applications
Is this finally the kind of digital assistant we've always wanted?
As the AI race unfolds, OpenAI keeps the lead and unveils GPT-4
Why it matters: OpenAI launched GPT-4 this week, an update to its popular language model and technology that aims to improve precision and is designed to act as an underlying engine for chatbots, search engines, online tutors, and more. GPT-4 is now available to paid subscribers and there's a waitlist to use the model via API. Furthermore, the AI race is on, with "AI startups" raising funds like there is no tomorrow and big tech companies like Google scrambling to make it known that they are not so far behind.
Microsoft's GPT-powered Bing Chat will call you a liar if you try to prove it is vulnerable
It also gets "very angry" when you call it by its internal codename Sydney