Meet Gemini (and say goodbye to Bard)

Howdy, wizards (DALL-E was feeling spooky today 😄)

I’m hitting the send button twice this week to promptly (pun intended) introduce you to the new kid on the LLM block: Gemini Ultra.

‍

What’s brewing in AI (mid-week update):

• Gemini Advanced is here
• Gemini Ultra first impressions
• The other top stories in AI this week

‍

Dario’s Picks: Gemini Ultra launch special

Meet Gemini (and say goodbye to Bard)

‍

‍

TLDR:

• Gemini Ultra 1.0 just launched  (Google’s long-awaited, top performing LLM).
• Bard rebranded to Gemini, available at https://gemini.google.com/app.
• Gemini Advanced: $20 paid tier for Gemini Ultra (2-month free trial).
• New Gemini app for Android, on iOS it will be part of the Google app.
• No API access to Gemini Ultra yet, but Google will have related announcements over the next weeks.
• See also Google’s blog post announcing the launch.

‍

Gemini Ultra is, in Google’s words, “far more capable at highly complex tasks with a range of applications like coding, logical reasoning, following nuanced instructions and creative collaboration, amongst many others”. It allows for longer prompts, improved context understanding and is multi-modal (can engage in chats about images you upload).

Gemini Advanced is the rebranded Bard chatbot, now powered by the latest Gemini Ultra model. It is also part of Google One, meaning subscribers will get 2TB of storage and the rest of Google’s One features.

Notably, it will soon also included Gemini in Google Workspace apps like Docs, Slides, Sheets and Meet, replacing the Duet AI brand which already had similar features inside these apps.

The pricing for Gemini Advanced is on the same level as ChatGPT Plus (20$/month). There’s also a 2-month free trial if you’re already a Google One subscriber. The free version of Gemini is still powered by Gemini Pro 1.0.

‍

‍

The AI Premium plan on Google One is the only one that includes access to Image source: Google.

The rebrand isn’t really surprising considering the mediocre performance of Bard compared to ChatGPT (particularly GPT-4). The company likely considers it easier to start “from scratch” than recovering customer perceptions of Bard as a second-rate chatbot.

Gemini Advanced will be available in 150 countries and only in English in this first version. The company lists Japanese and Korean as upcoming languages on the roadmap for the chatbot.

I’ve just subscribed to Gemini Ultra and will be using it extensively over the next weeks. I’m as excited as you are to see if it lives up to the hype and how it compares to ChatGPT. Stay tuned for my thoughts and comparisons.

‍

Why it matters: Since Bard was the second-biggest chatbot for productivity, Gemini is now ChatGPT’s largest competitor.

Gemini Ultra has been touted as Google’s very best LLM to date, reportedly outperforming GPT-4 on several benchmarks.

Since many people are already Google One users (like myself), upgrading to AI Premium is actually cheaper than paying for ChatGPT plus. So in this sense, Google has the upper hand on pricing, which might get a lot of new customers in the door.

Thus, OpenAI’s next move might hinge on the performance of the model and how well users receive the new chatbot. If Gemini Ultra performs well and/or wins market share quickly, it might push OpenAI to release new updates and models (like GPT-5) sooner. On the contrary, if it flops (like Bard arguably did), it’ll reinforce the number 1 position of OpenAI.

‍

‍

In Focus

Gemini Ultra first impression: “Meh”

‍

‍

Kristian from All About AI has done some quick tests on Ultra vs GPT-4 on some fairly simple tasks. He wasn’t too impressed with the new model, and even questioning whether or not his Gemini Advanced account had actually been updated with the new Ultra model.

‍

Some of his takeaways:

• It responds much faster than GPT-4.
• The user interface is great, including good code highlighting.
• GPT-4 was a bit better on simple tests for math and logic.
• Gemini Advanced managed make a simple snake game on the third attempt (one-prompt), which GPT-4 can also do.
• It occasionally responds “I’m a text-based AI, that’s outside of my capabilities” answers for questions it should be able to answer. This happened in the video with providing a code explanation and generating an image.

‍

My thoughts: I’ve played around with Gemini Advanced for a 15 minutes. While I wasn’t particularly impressed with the image generation, it worked. In other words, the issue above seems to happen only for some users.

However, I kept getting the “I’m just a language model, I can’t do that”-type of response A LOT, even for simple text-based tasks like “rewrite this paragraph”.. Let’s hope this is related to some early bugs that Google is currently fixing.

‍

‍

Featured GPT (Sponsored)

AI Beauty Assistant

‍

‍

A beauty assistant specialised in skincare and cosmetics. Customises advice to your needs, skin type, location and more.

‍

Bytes

General

• OpenAI is adding watermarks to DALL-E 3.
• Microsoft announces newsroom partnership with media company Semafor, amid NYT lawsuit.
• Dating app Bumble created an AI tool that removes 95% of spam and fake profiles.
• A company lost 25.6$ million to scammer recreating multiple coworkers with deepfake audio and video.
• An article about OnlyFakes, an underground website selling AI-generated fake IDs that can bypass crypto Know Your Customer (KYC) attempts.

New tools and features

• Roblox reveals in-game, real-time AI translation for 16 languages. Currently text based, but voice chat is on the roadmap.
• Amazon announces its AI shopping assistant Rufus. Launched to a small subset of customers, and rolling out gradually to all US customers in the coming weeks.
• The Promenade, an MMORPG video game/social network populated by AI agents, just raised $500k from a16z.

Research

• Tech students win the Vesuvius challenging, receiving $700k after reading ancient scroll with AI.
• 70% of Microsoft Copilot users report new AI features helping them work faster.
• Acoustic model fusion – a new technique from Apple for improved speech recognition.

Chips

• Meta is developing its proprietary AI chips to lessen dependence on Nvidia.
• Nvidia’s CEO says next big customers is nations seeking their own AI systems.
• Strong demand for Huawei AI chips (after Nvidia had to downgrade their exported chips) has the company prioritising AI over smartphones.

‍