- Everyday AI
- Posts
- Is Google Gemini a Failure? š¬
Is Google Gemini a Failure? š¬
š The truth about Google Gemini, AI translates brainwaves to text, Diagrams Show Me plugin review, and more!
Everyday AI
December 12, 2023
Outsmart The Future
Sup yāall,
If you didnāt already know, Google was kinda not telling the whole truth in its Gemini release. We kinda roasted them for it on todayās show.
What was your thought? After you vote, you can leave a comment. Weāll feature our favorite comments in tomorrowās newsletter.
āļø Jordan
Whatās your take on Googleās Gemini announcement and marketing? |
Today in Everyday AI
7 minute read
š Daily Podcast Episode: Googleās release of Gemini has not gone well. Does its marketing controversy already make it a failure? Give it a listen.
šµļøāāļø Fresh Finds: An instant thought capture app, AI made of human brain cells, and an AI Van Gogh museum. Read on for Fresh Finds.
š Byte Sized Daily AI News: AI translates brainwaves to text, Hallucinate is Dictonary.comās word of the year, and Microsoft turning to nuclear power for AI. For that and more, read on for Byte Sized News.
š AI In 5: Even if you don't use diagrams often, we have a plugin that can offer you a lot of benefits. See it here
š§ Learn & Leveraging AI: So whatās up with the whole Google Gemini released? Weāre breaking down what you need to know and why itās important. Keep reading for that!
ā©ļø Donāt miss out: Did you miss our last newsletter? We talked about Using AI automations for media marketing, General World Models announced, EU finalizes AI Act, and a ScholarAI ChatGPT plugin review. Check it here!
Google Gemini - ChatGPT killer or a marketing stunt? š¬
Well Google Gemini is FINALLY hereā¦and it hasnāt been great.
Google has been under fire for its Gemini release due to a demo video that misrepresented what its newest LLM was capable of.
Sorry to say but Google got so many things wrong with the marketing and launch. Yikes.
Is Gemini an actual ChatGPT killer or just a marketing stunt gone wrong? We're covering everything you need to know.
Join the conversation and ask Jordan any questions about Google Gemini.
Also on the pod today:
ā¢ How the Gemini demo was created š„
ā¢ Comparing ChatGPT to Gemini š„
ā¢ Why did Google release Gemini? š¤·
Itāll be worth your 48 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Upcoming Everyday AI Livestreams
Wednesday, December 13th at 7:30 am CST ā¬ļø
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā Ideaflow is an instant thought capture app, Intelogos collects performance data and generates suggestions that help your team grow, and MagicDocs helps organize and summarize your documents.
Science in AI ā An AI made from human brain cells (yes you read that right) learned to recognize a specific voice.
Social Media - Snapchat+ subscribers just got a new feature that allows you to create and send AI images.
Pop Culture ā An AI Van Gogh museum may be the answer to reviving art interest.
Read This ā Check out how AI can be used to stop the selling of counterfeit products.
1. AI Translates Brainwaves to Text š§
Scientists have developed an AI system, DeWave, that can translate brainwaves into written text. Using a sensor-filled helmet to record electrical brain activity, the team achieved around 40% accuracy, with recent improvements exceeding 60%. This non-invasive, portable technology, presented at the NeurIPS conference, offers promising applications for individuals who have lost speech due to conditions like stroke.
2. 'Hallucinate' is Dictionary.com's 2023 Word of the Year š
Dictionary.com has chosen 'hallucinate' as its 2023 Word of the Year, reflecting the growing influence of AI in public discourse. The word's second definition now includes AI's production of false information, mirroring its technological evolution similar to words like 'spam' and 'virus.' This choice highlights the complexities and limitations of AI, as seen in large language models like ChatGPT.
3. Microsoft's Nuclear Power Strategy for AI Operations š
Microsoft is exploring nuclear power as a solution for its escalating electricity needs, driven by its ventures into artificial intelligence and supercomputing. The tech giant is experimenting with generative AI to streamline the U.S. nuclear regulatory process, leveraging its investment in OpenAI. By training a large language model on nuclear regulatory documents, Microsoft aims to expedite the lengthy and costly approval process for nuclear projects.
4. Worldās First AI-Powered Political Campaign Bot
Shamaine Daniels' congressional campaign is leveraging Ashley, an AI-powered phone banker, to engage with voters. Unlike typical robocalls, Ashley offers personalized, real-time conversations using generative AI technology similar to ChatGPT. This innovative tool, developed by Civox, allows for tailored discussions based on voter profiles, enhancing campaign outreach. While Ashley's use raises concerns about disinformation and election integrity, Civox has implemented measures like a robotic voice and AI disclosure.
5. Sports Illustrated's AI Scandal Leads to CEO's Firing š°
The Arena Group, publisher of Sports Illustrated, has fired CEO Ross Levinsohn following a scandal involving AI-generated articles and fake authors. The controversy erupted after an investigative report revealed that Sports Illustrated published content by non-existent writers with AI-generated headshots and biographies. The board's decision to terminate Levinsohn aims to improve operational efficiency and revenue.
Diagrams Show Me ChatGPT Plugin Review! š
The Diagrams Show Me ChatGPT Plugin is amazing!
Even if you don't use diagrams often, there's still a lot of benefits you can get out of this plugin.
We're showing you how to use it, why it's so good, and some practical applications for it to accelerate your learning.
Check out today's AI in 5.
š¤·āāļø Whatās Going On and Why It Matters:
Wanna know whatās wrong with the state of AI right now?
(Well, itās actually a lot.)
But one BIG thing that weāve seen recently: everyone on the internet blindly copying and pasting marketing and messaging from big tech companies when they release a new AI product.
It kinda grinds our gears.
Thatās why we had to take a deep dive today on the real scoop behind Googleās Gemini release.
Google officially unveiled its Gemini models last week, making way for Gemini to replace the PaLM 2 model in its Bard chat and other AI offerings.
With the snazzy marketing video below, Gemini looked like the next best thing since sliced bread. Even better than that.
Let's go hands-on with #GeminiAI.
Our newest AI model can reason across different types of inputs and outputs ā like images and text. See Gemini's multimodal reasoning capabilities in action ā
ā Google (@Google)
6:29 PM ā¢ Dec 6, 2023
But.
Buuuuuuuuuuuut. It was all faked.
Yet, the media and social media influencers started labeling Gemini as the next big thing right away, not even attempting to read the fine print or investigate it further.
Thatās why Jordan had to cut through the fluff and bring the real story about Google Gemini.
Todayās episode is definitely a must-listen, but hereās the low-down:
Introduction to Google's Gemini Model
Deceptive marketing tactics and misrepresentations
Impact on Google's stock value and credibility
Assessing Gemini's Performance and Functionality
Examination of multimodal capabilities
Highlighting the need for accurate product testing
Emphasizing the impact of misrepresentation on trust and credibility
And a ton more.
So, letās break it down and give you the real truth, the whole truth, and nothing but the truth.
š¦¾How You Can Leverage:
Hereās where we separate the marketing from the truth.
And weāre pulling no punches here, shorties.
Our hot take: when big tech companies like Google tell half-truths in their marketing, it hurts the whole GenAI industry.
Trust and transparency are paramount in all things AI. And Google dropped the ball.
Maybe your company was on the fence about leveraging GenAI in your day-to-day, and this Google SNAFU pushed it to the āGenAI isnāt for usā side of the fence.
Also worth reiterating ā there are three different models.
Ultra ā the most powerful version that Google ādemoedā isnāt even available until next year.
Pro ā This version is available now inside Google Bard, and itās very impressive.
Nano ā This version is coming to Googleās smartphone lineup, and is compact enough to run locally on a device.
With that out of the way, letās cut through the PR and talk about what really happened in Googleās Gemini rollout, and what it means for the rest of us.
1 ā Nothing was real-timeļø š³
Hot dang. š«£
That super impressive marketing video where the user and Google Gemini had a real-time conversation, showing off impressive abilities in speech, understanding and reasoning?
All faked. Not real-time. Created after the fact.
What did they really do? They recorded a video, then took many screenshots of the video. Then, they fed those screenshots into Gemini to get the responses. Oh, but they also needed to input in-depth text prompts (that were NOT in the marketing video dialogue, obviously) to get the results you heard in the video.
Try this: Hey Google, how about we get the full, real truth here? Because even after getting called out on faking the video, execs keep saying the demo video is ācompletely real.ā
Nope. Not at all. Not only was the video produced AFTER the fact, but they had to include additional (multi-step!) text prompting to generate the results that were seemingly marketed as real-time.
Go read the Google blog post on how the video was made, and youāll notice that not only is the marketing video not at all real, but Google even had to input extra prompting (that was not included in the video dialogue) in order to produce the results.
Oh, and this āfeatā of Gemini is already very possible in the already available GPT4. š
2 ā Numbers donāt lie š¢
Ooof. Another Google swing, another Google miss.
Where do we start here?
In Googleās Gemini report, they share some not-super-impressive statistics.
If you want the full numbers breakdown, make sure to re-watch (or re-listen to) todayās episode. But, the numbers arenāt nearly as impressive as you might think.
Try this:
Google touted that Gemini outranked GPT4 in 30 out of 32 benchmarks popularly used in testing LLMs.
Kinda true.
But, thatās with the Ultra model that isnāt even out yet and the public canāt test at-large.
So, what about the āProā model thatās available now inside Bard? Well, GPT4 takes it to school and grabs its lunch money. š
So, Google hyped up the Ultra model that we canāt use or test, but the Pro model is significantly worse than the GPT4 model that will soon be 1 year old.
3 ā Todayās reality is what matters š”
Hereās where we think Google REALLY dropped the ball.
Yeah, their marketing video wasnāt factual and more hype than substance. Pretty normal for tech companies.
But what has us really scratching our heads is why Google released a (subpar at best) Gemini Pro model for the world to see and use right now.
Google essentially said, āHEY EVERYONE, COME TRY GEMINIā
But all thatās available is the Gemini Pro variety, which performs more similarly to GPT3.5 than GPT4. In other words, why draw attention to a model thatāsā¦. not very good?
Facepalm.
Try this: Go find out for yourself.
Like we do and encourage others to do, go test out the Gemini Pro version of Bard thatās live right now.
Now This ā¦
Do you plan on using Google Gemini?
Hit us with a reply and let us know!
Reply