|
On Monday, May 13, OpenAI launched its newest model, GPT40, the day before Google’s annual I/O event. With impressive low latency multi-modal capabilities, the system displayed extraordinary understanding and vocal nuance. However, one of its voices was eerily similar to Scarlett Johansson’s character in the 2014 movie ‘Her.’ This led to a lawsuit threat from Johansson and her legal team a week later. OpenAI protested, claiming the similarity was unintentional, but their timing suggested otherwise.
Johannson’s statement revealed that OpenAI had contacted the actress the weekend before launch to ask if she’d change her previous ‘no’ to a ‘yes.’ Although the GPT4o voice belonged to another actress, this revelation didn’t help OpenAI’s optics. Sam Altman’s decision to post the word ‘Her’ on X minutes after GPT4o’s reveal didn’t help either. The result? The voice model was pulled from the product indefinitely, proving that ‘OpenAI is nothing without its drama.’
Following what appeared to be a successful launch, OpenAI had a terrible Tuesday. Co-founder and creative genius Ilya Sutskever and the company’s safety team co-head Jan Leike resigned. CEO Altman paid tribute to his “dear friend” Ilya, while Leike fired a volley of tweets on X accusing his former company of prioritizing “shiny toys” over AI safety. Altman responded with a lukewarm tweet of super appreciation for the work Leike had done at the company.
Update:
As we were finalizing this issue, a video emerged from Vivatech Paris in which Romain Huet, Head of Developer Experience at OpenAI, demoed some of ChatGPT-4o's more advanced capabilities.
Huet showed off real-time language translation, landmark identification from a simple sketch, map reading (with travel directions) from a map held up to the screen, real-time coding assistance, and agent-based API calling functions to a map app. Huet then stated that the model offers a complete agent-based toolkit with assistant functions, conversational history, and the ability to upload and retrieve up to 10,000 files.
Finally, Huet demoed the model sampling frames from a generated Sora video of 19th-century Paris, provided a 15-second sample of his voice, and told the model to create a voiceover narration that he seamlessly switched across multiple languages in near real-time.
It was an incredibly impressive display and showed OpenAI's new ‘Omni’ models moving ahead of the competition with truly game-changing multi-modal capabilities.
Google Bytes Back
Never one to go unnoticed at a party, Google’s I/O keynote kicked off with a surreal DJ set from musician Marc Rebillet, which was either groundbreaking or peak AI cringe, depending on your viewpoint.
Google’s week was a whirlwind of AI razzle-dazzle. If there was any doubt about their commitment to AI, they erased it, using the term 120 times during the evening keynote! But, to give them their due, they did have a lot of AI news to share, announcing a truckload of products, including:
One of the standouts was its new Veo video model, which looks like a legitimate competitor to OpenAI’s Sora. With the ability to produce 1080p videos that go beyond a minute and a wide range of visual styles, it has the potential to push boundaries.
Microsoft Build 2024
Microsoft has been a big mover in the AI space for some time, and its latest AI updates were designed to remind everyone of its AI boss status!
- First, brace yourselves for updates to Copilot Studio with custom co-pilots. These allow you to create AI agents to automate your business workflows.
- Next, extend CoPilot with GitHub extensions. Add knowledge with API plugins, graph connectors, prompts, flows, actions and more.
- Then, experience brand-new hardware like CoPilot+ PCs with AI-optimised silicon and a Recall feature, which lets you remember everything you’ve done on your PC. Creepy or cool, you decide!
Microsoft also flaunted its partnership with NVIDIA, name-dropped ChatGPT, and extended its team-up with Hugging Face, bringing their models to Azure AI Studio. Cognition AI’s hot new startup, Devin, which recently emerged from stealth, will also be powered by Azure, ensuring intelligent agents get smarter.
Alongside new Azure VMs powered by Cobalt 100 processors, which aim to make data centers greener, AMD’s ND MI300X series processors are now in Azure, perfect for enterprises tackling massive AI computing tasks.
Enjoy the ride, devs. Microsoft’s got your back.
Featured Blogposts
Need Extra Dev Superpowers? We’ve got 65!
So, devs, we all know ChatGPT is our new superhero sidekick—the Robin to our Batman! The proper prompts draw out its amazing capabilities. But when you're knee-deep in code and coffee, who has time to find the best!!?
We do! ;)
Forget banging your head against your mechanical keyboard—our comprehensive guide has 65 prompt power-ups tailored just for you. Like a Mario magic mushroom, these prompts will supercharge your workflows, help you create slick code, optimize for speedy performance, and boost SEO.
Click our guide here to power up your prompt game and find extra sanity.
AI Agents: The Future of AI?
The power of AI Agents lies in their ability to perform complex multi-step tasks. AI Agents iterate, refine, and enhance their performance using reasoning models to reach the final goal.
If GPT-3.5 is 48.1% accurate, adding an agent loop skyrockets that to 95.1%, making GPT-4's 67% look stuck in dial-up!
AI Agents are here, and they’ll soon offer you an AI development, design, finance, and executive team at a fraction of the cost. No off days, long holidays, or snarky letters of resignation! Instead, a future where small groups of talented humans partner with AI to offer amazing goods and services—think R2-D2 helping Luke destroy the Death Star!
If you’re curious about how this ALL works, visit our main site for a deep dive!
MLOps & LLMOps Unlocked:
If wrangling text data for LLM training has you feeling like a budding linguist deciphering ancient hieroglyphics, don't despair. This isn't your average piece of AI-generated fluff. In this three-part series, we dig down from the nitty to the gritty, showing you how to wield BigQuery like an SQL samurai, tame the tangles of your messy text datasets, and fine-tune your models with the finesse of a seasoned sensei. Prepare to climb to the top of the LLM game and ascend the levels to transform your models from jagged rocks into polished jade.
|
