- AI Tool Report
- Posts
- 🍎 Apple exposes major AI flaw
TOGETHER WITH GUIDDE
Welcome to AI Tool Report!
Tuesday’s top story: In a new self-conducted study, Apple researchers have discovered a major flaw with all major AI models.
🌤️ This Morning on AI Tool Report
🍎 Apple exposes major AI flaw
🎥 How to create video guides in less than an hour, with no skills
💥 Google goes nuclear
📰 How to get key takeaways from your newsletters
🧐 How to understand your competition using ChatGPT
📽️ Adobe’s new AI video generator
🫨 Microsoft loses key AI VP to OpenAI
Read Time: 5 minutes
FACT OF THE DAY
🤔 45% of people use AI to respond to messages via text or email, 43% use it to answer monetary questions, 38% use it to plan their travel itineraries, 30% use it to prepare for a job interview, and 25% use it to write social media posts.
STOCK MARKETS
⬆️ Stocks soar higher to start the week, with NVIDIA out-performing the broader market, continuing the momentum shown last week. The chip maker is under 2% from the all-time highs, previously made in late June. In other news, Taiwan Semiconductor (TSM) is predicted to generate $23B in Q3 revenue, as the AI demand grows. Learn more.
— — — — — — —
PERFORMANCE
Our Report: Apple tested over 20 Large Language Models (LLMs)—including OpenAI's o1 and GPT-4o, Google's Gemma 2, and Meta's Llama 3—to see if they were capable of "true logical reasoning," or whether their ‘intelligence’ was a result of "sophisticated pattern matching" and the results revealed some major weaknesses.
🔑 Key Points:
LLM’s reasoning abilities are usually tested on the popular benchmark test—GSM8K—but there’s a probability that the LLMs can only answer questions correctly because they’ve been pre-trained on the answers.
Apple’s new benchmark—GSM-Symbolic—tested this by changing variables in the questions (eg. adding irrelevant information/changing names or numbers) and found every LLM dropped in performance.
As a result, they believe there is “no formal reasoning” with LLMs, “their behavior is better explained by sophisticated pattern matching” as even something small, like changing a name, degraded performance by 10%.
🤔 Why you should care: If LLMs can only rely on sophisticated pattern matching, rather than genuine logical reasoning, it means they can’t be relied on for AI applications that require consistent, accurate reasoning to help with real-world situations and environments, although it’ll be interesting to see how Apple, with its own AI models, and a major competitor of Google, Meta, and OpenAI (despite their new partnership), will respond to its own disparaging study.
— — — — — — —
TOGETHER WITH GUIDDE
FAQs, training materials, onboarding docs, how-to guides, and feature notes take weeks to write. They always need updating, are often off-brand, and take whole teams of people to create.
But no one reads them.
What if you could create, edit, publish, and update brand-consistent, engaging video guides in less than one hour?
Welcome to Guidde, your all-in-one AI platform for effortless video guide creation for any organization, big or small.
Our easy-to-use features mean you can:
Quickly whip up videos
Edit in real-time
Share with anyone, anywhere
TRENDING TOOLS
Brilliant: Future-proof your skills in minutes a day with a library of quick, interactive lessons in AI, programming, logic, data science, and more. Try it for nothing for 30 days ⭐⭐⭐⭐⭐ / 5 (G2)
illumidesk generates eLearning Lessons with AI
Kaiber is an AI video-generation tool
— — — — — — —
POWER
Our Report: Google has signed a world-first agreement to use carbon-free nuclear power produced by seven new Small Modular Reactors (SMRs)---which are more cost-effective and efficient than traditional, commercial-scale nuclear reactors—that will be built by nuclear power start-up, Kairos Power, to power its AI data centers, “accelerate the clean energy transition across the US”, and “unlock the full potential of AI for everyone.”
🔑 Key Points:
Google is backing the construction of the seven SMRs, which are expected to deliver 500 Megawatts of carbon-free electricity, with the first scheduled to come online by 2030, and the rest coming by 2035.
Although their plans haven’t yet been approved by the US Nuclear Regulatory Commission and local agencies, Kairos Power was given a permit (the first in over 50 years) allowing them to build a new type of reactor.
Reports say data centers currently consume around 3% of global electricity, with consumption expected to double by the end of 2030 as AI advances, so more electricity is needed to power them and keep them cool.
🤔 Why you should care: Although Google didn’t reveal any monetary details about the agreement or specify where the seven SMRs will be built, it's following in the footsteps of Microsoft and Amazon, as last month Microsoft announced it was partnering with Constellation Energy to restart a reactor at the infamous Three Mile Island site (which was shuttered in 2019), and Amazon confirmed, in March, that it was building a 1,200-acre (486-hectare) nuclear-powered data center and connecting it to a nuclear plant in Pennsylvania to power their AI operations for the next 20+ years.
— — — — — — —
TOGETHER WITH MECO
Meco, the #1 app for reading newsletters, just released an AI feature that gives the key takeaways from your newsletters in record time.
Just tell Meco your favorite newsletters and they’ll curate a daily 5-10 minute personalized audio briefing — ideal for quick, on-the-go listening, even when you're short on time.
⚡📲The app is packed with loads more features to supercharge your reading:
Connect your inboxes (Gmail/Outlook) to declutter them & consolidate newsletters
Offline reading, custom themes, highlight text, and bookmark links for later
Get AI text summaries for newsletters to get the key details in seconds
PROMPT ENGINEERING
— — — — — — —
Tuesday’s Prompt: How to understand your competition using ChatGPT
Type this prompt into ChatGPT:
Help me understand the regulatory landscape that affects our competition
Results: After typing this prompt, you will get an analysis of the regulations that affect your competition.
P.S. Use the Prompt Engineer GPT by AI Tool report to 10x your prompts.
It’s not too late! Join the AI Reports AI Skill Sprint on Skool and master 6 crucial AI skills in just 6 weeks…
✅ What You'll Learn: Course 3 is with no-code, automation specialist, Grant Hushek (founder of GrantBot Process Consulting) who will take you through topics related to operational automation, including the key differences between automation and AI automation and how to automate customer support processes, CRM updates, content idea generation, and more…
🫱🏻🫲🏻 Connect with Grant here
BREAKING NEWS
— — — — — — —
VIDEO
Adobe has launched its AI video generator, called Firefly Video Model (which they initially announced earlier this year), that will allow creators to generate videos from still images and text prompts.
They also launched Generative Extend (in beta for Premiere Pro users) which can be used to extend the end or beginning of footage or make adjustments mid-shot, like correcting unexpected movements.
The max length of Adobe’s AI videos is just five seconds (OpenAI’s Sora can generate clips up to 60 seconds long) and it takes around 90 seconds to generate, although Adobe is working on a “turbo mode” to reduce that.
— — — — — — —
INTERNAL POLITICS
Microsoft has lost its VP of GenAI research, Sebastien Bubeck, to OpenAI, although there is seemingly no bad blood, as they want to maintain their “relationship through his work with OpenAI.”
Bubeck was instrumental in developing Microsoft's Phi models—a series of small language models—so many believe he’s jumped ship to work on the efficiency of OpenAI’s smaller AI models.
Finally some good news for OpenAI, after recently losing several key people (notably 7 of the 11 original founders), and despite Bubeck’s departure, his team, who developed the Phi models will remain at Microsoft.
RECOMMENDED RESOURCES
🕊️ | 🕊️ | 🕊️ |
We read your emails, comments, and poll replies daily.
Hit reply and tell us what you want more of!
Got a friend who needs to learn more about AI? Sign them up to the AI Tool Report, here.
Until next time, Martin & Liam.
P.S. Don’t forget, you can unsubscribe if you don’t want us to land in your inbox anymore.
What did you think of this edition? |