The truth about Gemini

This week's biggest AI news

Happy Friday! Grok (the chatbot made by Elon Musk’s xAI) is officially available to X’s Premium Plus subscribers in the US. The second I got access, I asked Grok to roast me based on my tweets. Apparently, I should be known as “Matt ‘I’m-obsessed-with-AI’ Wolfe” and should consider having “a real conversation without mentioning AI.” Honestly…it’s not wrong.

P.S. I want to include more of what you want in Future Tools! Tell me your thoughts here.

Gemini Is Here (and Already Stirring the Pot)

Source: Google

This week, Google launched its flagship generative AI model Gemini—and it’s been a whirlwind two days of industry response.

Gemini is a natively multimodal AI model, meaning it was trained on different types of information (including text, code, audio, image, and video). But Gemini’s other big differentiator? It’s actually a set of AI models, not just one:

  • Gemini Nano is small enough to run on mobile phones, like Google’s Pixel 8, the first smartphone with on-device AI.

  • Gemini Pro now powers the latest version of Google’s Bard chatbot and will soon be able to integrate with other Google products. Developers and enterprise customers will be able to access Gemini Pro via an API on December 13.

  • Gemini Ultra has been the focal point of the launch—but it hasn’t exactly launched yet. As the largest, most capable version, Ultra apparently “exceeds current state-of-the-art results on 30 of the 32 widely-used academic benchmarks” used to assess LLMs. Ultra and Bard Advanced—powered by Ultra—are set to roll out next year.

How does Gemini compare to GPT-4? Google anticipated this comparison—and prepared marketing materials accordingly:

  • When compared to GPT-4, Gemini Ultra reportedly excels in Python coding tasks, grade-school math, reading comprehension and arithmetic.

  • Google says that Ultra is the first model to outperform human experts on MMLU (massive multitask language understanding), a test of a combination of 57 subjects.

  • Google’s launch video for Gemini demos the model. As a user draws a duck, then holds up a rubber duck, Gemini seems to be comprehending the duck’s transformation and responding in real-time. And as the user draws a guitar, Gemini seemingly generates a guitar sound instantly.

But comparison comes with caveats. First, the stats compare Gemini to the older GPT-4, not GPT-4 Turbo (which powers ChatGPT Plus and Bing). Second, the subject of comparison is Gemini Ultra, which hasn’t launched yet. Gemini Pro, the model currently powering Bard, only wins out against GPT-3.5 in benchmark assessments.

And the big disclaimer? In my first video about Gemini, I was seriously impressed by the demo video. But today, news broke that the video was misleading, to say the least:

  • According to a Google Developer blog (conveniently hidden in the video’s description), Gemini was actually responding to “a series of images,” not video, and its seemingly real-time responses were just a sequence of responses to those static images. Plus, the model didn’t actually generate the guitar sound…it just searched for the music.

Sure, there was a tiny disclaimer in the beginning of the video that “sequences [are] shortened throughout,” but much of the industry agrees: Google wasn’t entirely honest with their promotion of Gemini.

My takeaway? Google wanted to top OpenAI so badly, they embellished Gemini’s capabilities and misrepresented the tech. Personally, I think it’s a bit of a stain on their reputation now.

The All-in-One Password Manager for Teams

AI tools make us all more productive—but even the most powerful tools won’t make up for time lost to the “reset my password” cycle.

73% of employees waste 15+ minutes per day attempting to access company infrastructure. That’s why NordPass built the all-in-one password manager for teams. With NordPass, you can:

  • Generate strong passwords and store them all in one secure place.

  • Instantly and securely share passwords, passkeys, and other sensitive data with colleagues.

  • Identify breaches early before they harm your business.

  • Autofill passwords, company credit cards, and more across all devices.

Secure your business effortlessly with a 3-month NordPass trial! Use "futuretools" activation code at Limited time offer!

The Battle of Open Source Vs. Proprietary AI

Source: Yahoo Finance

Meta, IBM, and over 50 other organizations announced the formation of the AI Alliance, a consortium advocating for open-science AI (AKA transparent innovation across research, building and adoption).  

Why open source? “Frankly, we’ve been a little bit unsatisfied with the overall debate and the discussions on AI over the last year,” says Darío Gil, senior VP at IBM. He claims that the closed-source models of big AI companies doesn’t “reflect the diversity of the ecosystem.” AI Alliance members believe in a distributed approach to AI innovation—Gil explains that this is “much more resilient, because no given institution can derail the success of the open engine.”

So who’s in (and who’s out)? The AI Alliance includes high-profile organizations: universities like Harvard and Yale, big names in AI like Stability AI and AMD, and big tech companies like Intel, Oracle, and Dell… 

  • but critically excludes OpenAI, Anthropic, Microsoft, Google, and Nvidia. These AI heavyweights are mostly in favor of closed-source, or proprietary, AI systems. 

Why proprietary? Commercial incentives are undoubtedly playing into this stance. But representatives of these companies have also expressed fear of the security risks that could come with open-sourcing.

Bottom line: Critics argue that without big names like OpenAI and Google, the AI Alliance lacks the participation needed to achieve its goals. As AI companies race to convince policymakers on their stance, the divide between between the open-source and proprietary camps will only grow stronger.

AMD Takes on Nvidia

Semiconductor giant Advanced Micro Devices Inc. (AMD) just launched a new chip. Has Nvidia met its match?

In a nutshell:

  • AMD’s new Instinct MI300X chip was designed for speedy data transfer—specifically for larger AI models—featuring 192GB of super-fast memory called HBM3.

  • AMD is positioning the MI300X as “the most advanced AI accelerator in the industry”, claiming it even outperforms Nvidia’s state-of-the-art H100 processor.

  • While the chip’s price tag hasn’t been revealed yet, AMD claims it will be cheaper to buy and operate than Nvidia’s (which cost $40,000 a pop).

Competition heats up: Big Tech welcomed the news with open arms. Meta and Microsoft (the two largest buyers of Nvidia H100 chips in 2023) already said that they’ll buy the Instinct MI300X to power some of their AI workloads. It’s clear: companies are hungry to find cheaper alternatives to Nvidia’s coveted GPUs. The Instinct MI300X will start shipping early next year.

Bonus Story:

OneLLM understands eight different types of data… including brain activity. This new AI model is giving “multimodality” a whole new meaning. Interested in how it works? Check out this paper.

  • Alibaba’s new image-to-video model went viral for scraping videos from famous TikTokers.

  • Meta launched a free, standalone image generator called Imagine with Meta.

  • Google researchers cracked ChatGPT by getting it to spit out private training data.

  • Microsoft unveiled Deep Search, a new Bing feature that makes prompting AI easier. 

  • Runway partnered with Getty Images to launch a powerful new AI video model for enterprises.

More important AI news: Dive deeper into this week’s hottest AI news stories (because yes, there are even more) in my latest YouTube video:

AI video is on fire: There are some wild advancements happening in the world of AI video right now. Check out my video unpacking it all:

Everything changed: Celebrating ChatGPT’s first birthday, here’s a recap of the chatbot’s crazy fast evolution.

And there you have it! What do you think the Gemini launch means for the future of AI models? Shoot me a reply with your thoughts. Catch you back here on Wednesday!

—Matt (

P.S. This newsletter is 100% written by a human. Okay, maybe 96%.