- Everyday AI
- Posts
- Are LLMs ACTUALLY connected to the Internet? š¤
Are LLMs ACTUALLY connected to the Internet? š¤
LLM showdown, Microsoft invests $1.5B in UAE, Google DeepMindās $100B AI plan, and more!
Everyday AI
April 16, 2024
š Subscribe Here | š£ Hire Us To Speak | š¤ Partner with Us | š¤ Grow with GenAI
Outsmart The Future
Today in Everyday AI
6 minute read
š Daily Podcast Episode: Are LLMs like ChatGPT, Gemini, and Copilot actually connected to the internet? We put them to the test in a showdown. Give it a listen.
šµļøāāļø Fresh Finds: Meta investigating explicit AI-generated images of public figures, Baiduās Ernie Bot reaches 200M users and using multiple chatbots at once. Read on for Fresh Finds.
š Byte Sized Daily AI News: Microsoft invests $1.5B in UAE, Google DeepMindās $100B AI plan and LinkedInās top companies for AI growth. For that and more, read on for Byte Sized News.
š AI In 5: We just found an AI-powered software that lets you use multiple LLMs at once! Weāre showing you how it works and why you might wanna use it. See it here
š§ Learn & Leveraging AI: So are LLMs connected to the internet like they say they are? Weāre breaking down how ChatGPT, Gemini and Copilot performed. Keep reading for that!
ā©ļø Donāt miss out: Did you miss our last newsletter? We talked about Adobe using AI images for data training, Googleās new AI features, Apple making AI local to iPhone and the UK to regulate AI models. Check it here!
Are LLMs Actually Connected To The Internet? š¤
Ready for a little LLM secret?
Big tech companies are lying to us.
They tell us their models are connected to the internet in real time. Are they really? Not really.
We're doing an LLM breakdown, as we compare the internet connectivity of popular large language models like ChatGPT, Gemini and Copilot.
Join the conversation and ask Jordan questions on LLMs here.
Also on the pod today:
ā¢ Testing Large Language Models š§āš¬ļø
ā¢ Implications for Future Use of LLMs š ļø
ā¢ Using Large Language Models in Business š¼
Itāll be worth your 34 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Upcoming Everyday AI Livestreams
Wednesday, April 17th at 7:30 am CST ā¬ļø
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā SEOmaker gives AI SEO insights, Poe lets you use multiple chatbots at once, and Muzaic API lets you add music anywhere.
Trending in AI ā Meta is investigating how Instagram in India and Facebook in the U.S. handled explicit AI-generated images of public figures.
AI Chatbots - Baidu says its chatbot Ernie Bot has 200 million users.
Big Tech ā AI leaders canāt agree on ways to test how responsible AI models are.
Future of Work ā AI is causing intelligence agencies to rethink how they work.
1. Microsoft to Invest $1.5 Billion in UAE AI Firm G42 š¦šŖ
Microsoft is dropping $1.5 billion into G42, an AI firm in the UAE. Brad Smith, Microsoftās President, is joining G42ā²s board of directors. Thereās a groundbreaking commitment in place to adhere to top-tier standards in AI development, with the US and the UAE government giving it the nod of approval. And thatās not all - whispers of alleged ties, fiery denials, and a jaw-dropping $1 billion developer fund promise to keep the tech world on its toes.
2. Google DeepMind Unveils Plans for $100 Billion AI Investment š¤
Demis Hassabis, the CEO of Google's DeepMind, made a groundbreaking announcement about the tech giant's ambitious venture to invest over $100 billion in advancing artificial intelligence technology. This revelation not only solidifies Google's dedication to AI progress but also escalates the ongoing investment frenzy in Silicon Valley. The mention of a potential collaboration on a cutting-edge supercomputer dubbed "Stargate" between Microsoft and OpenAI further fuels speculation about the future landscape of AI innovation.
3. LinkedIn Unveils Top Companies for AI Growth in 2024 š
LinkedIn's annual ranking reveals JPMorgan Chase & Co. as the top company, leading in AI initiatives with the AI Maker Space facility at Carnegie Mellon University. Amazon follows closely with the Amazon AI Ready initiative aiming to train 2 million employees in AI skills by next year. Notably, these companies are providing tools and reskilling opportunities to enhance generative AI capabilities among their workforce.
4. Amazon Music Unveils Maestro: Your Personal DJ š§
Amazon Music introduces Maestro, an AI playlist generator allowing users to create playlists through spoken or written prompts, including emojis. This innovative feature aims to enhance user experience and engagement by providing personalized playlists based on user input. Customers can expect a fun and interactive way to curate music tailored to their activities, moods, and preferences.
5. NSA Drops Major Intel on AI Security Measures šŗšø
The National Security Agency (NSA) is spilling the tea on deploying AI systems securely with a fresh Cybersecurity Information Sheet (CSI). This guidance aims to support National Security System owners and Defense Industrial Base companies in operating AI systems developed by external entities. The release marks a significant milestone in the NSA's efforts to enhance the security of AI systems, especially in high-threat and high-value environments.
This LLM Gem could change the game!
We just found an AI tool gem.
ChatHub lets you use multiple LLMs at once. Weāre talking ChatGPT, Claude 2, Perplexity and more!
Weāre showing you how it works.
Check out today's AI in 5.
š¦¾How You Can Leverage:
Shocker ā big tech companies are lying to us.
(Not really a shocker, we know.)
Donāt worry, youāve got a friend in AI who will tell you the truth.
(Thatās us.)
Something that many of us take for granted is not just understanding how large language models work, but also the ways they retrieve up-to-date info from the internet.
So, we tackled this topic today, and itās 32 minutes thatāll be worth your time.
We took a look at how ChatGPT, Google Gemini and Microsoft Copilot all access (or donāt access) the internet.
And before you get your pythons in a bunch, weāre intentionally not including some of the bigger names here in Perplexity and Claude.
Mainly because Perplexity is an answers engine powered by GPT and Claude, and Claudeās not being included because itās not internet-connected at all.
With that outta the way, letās get the skinny on how models are connecting to the internet, and uncover why Big Tech may not be telling us the whole truth.
And in todayās 1-2-3 breakdown, weāre not just recapping the highlights of todayās show, but weāre also diving deeper and wider to give you more context on why this matters.
1 ā Hallucinations still exist with āonline LLMsā š¤¦
Hereās some simple facts ā large language models have knowledge cutoffs, or dates that their training data is current through.
So, if you ask an LLM like Gemini or ChatGPT or Copilot information about what happened 2 months ago, that answer does not exist in its training data.
Models may say they donāt know, or just make crap up. (Usually itās the latter.)
Big Tech has seemingly solved that by connecting their models to real-time search, with ChatGPT and the GPT-based Microsoft Copilot both connected to Bing search and Gemini (sometimes) connecting to Google search.
As we showed live today: just because a model is connected to the internet, doesnāt mean itās going to be foolproof. So donāt fall for the marketing.
Try this:
Hereās a mind-bender for your Tuesday reading.
As important as real-time internet access is for large language models, they may actually be destroying the internet. In a (kinda?) good way.
2 ā And the winner is š
In our very unofficial/not scientific/unscripted live teardown, ChatGPT with a web-enabled GPT came out the winner.
But the reality is, ChatGPTās Browse with Bing, ChatGPT with a GPT, Google Gemini and Microsoft Bing all struggled to get the truth right.
So are companies lying to us?
Weād say so, yes.
Companies like Google, Microsoft and OpenAI tell us in their marketing that their models are using āreal-time data' and access ācurrent information.ā
Butā¦. Theyāre not.
Even in our short test when we ask for up-to-date information on the markets, these internet-connected models may bring in information that is weeks old.
Try this:
Knowing how to use a LLM is essential for your career and your company. If you donāt know what youāre doing, youāre leaving a lot to chance. (Hint: that wonāt end well.)
Get the basics covered with our free PPP class.
What are you waiting for?š
3 ā How LLMs and the internet will co-exist š¤
Hereās the reality ā right now, LLMs canāt be updated fast enough.
The whole Reinforcement Learning from Human Feedback and model fine-tuning are both time-consuming and expensive.
So for the foreseeable future, models will continue to be pretty outdated and spit out falsehoods left and right.
So theyāll need access to the internet.
Hereās how we see it working.
Try this:
Our prediction is this ā the future of using LLMs will look more like either Perplexity or Googleās SGE right now than using ChatGPT or Google Gemini.
Weāre not talking from a UI/UX perspective, weāre talking about under the hood.
We (kinda jokingly) dubbed the process of calling the internet while using a model as āMini-RAGā and we see this as the future.
Picture this: when you start a new chat in ChatGPT, imagine how useful it would be if the model FIRST did a search/citation of 20+ sources like Perplexity, but then also allowed you to toggle sources on and off depending on their relevancy and also gave you an option to input URLs or documents of your own before getting your results back.
An extra step or two? Sure.
Cutting down on LLM hallucinations and giving you WAY more usable outputs in the end in a faster fashion?
Absofrigginlutely.
We wanna know below! š
We took a little different approach with todayās newsletter and our livestream. (Donāt worry, #HotTakeTuesday will be back!)
In our livestream, we showed you more real-time results and comparisons, and took a bit of a āfuturistā role in the newsletter.
What are your thoughts?Vote to see live results! |
ā
Numbers to watch
55%
55% of organizations reported using AI in 2023, up from 50% in 2022 and 20% in 2017.
Reply