đ¤ w/146: AI Lessons From Harvard Business School
Plus: Apple's long game AI strategy, UK's AI Summit, the week's top AI tech stories and Sam Bankman-Fried is found guilty of crypto fraud.
w/Wiser! #146 - 5th Nov 2023
In this issue:
Harvard Business School Study tells us what we need to know about using AI in the workplace
UK hosts first of a kind global summit on AI
Appleâs AI long game.
Plus the top stories on whatâs going on in AI and emerging technologies shaping tomorrowâs digital world and your workplace.
w/WiserWords!
AI Lessons From Harvard Business School
I get asked a lot of questions about using AI. By far the biggest category is the âis it worth it and can I rely on it ?â group of questions. The short answer is âyes, you can.â But itâs a conditional answer because the slightly longer and more correct answer is âyes, you can, but be careful, know what itâs good at, what itâs not and remember that the buck stops with you!â
To support this advice I lean on the data from a recent working paper by Harvard Business School. The study involved 758 consultants from Boston Consulting Group and itâs purpose was to look at how to best use AI tools in the workplace.
The headline finding is when you train people to use AI, they produce higher quality results faster than people with no training and people who are not using AI at all.
BUT (thereâs a âbutâ)âŚthis only applied for tasks within the AIâs core capabilities. Which makes sense when you think about it. Because when you give AI a task itâs not trained to do, the human wins hands down, every time. Itâs like using a calculator to write words. After âhello,â âbill,â and âboobiesâ youâre running out of options.
For the study, Harvard used GPT-4 for the test, which involved splitting the BCG consultants into 3 groups; those that had been given âpromptâ training, those that hadnât but were left to use the AI anyway, and a third group that had no access to ChatGPT at all.
When given tasks to do that were within ChatGPTâs range of capabilities, the study found:
Consultants using AI produced higher quality results.
They were also faster and more productive.
Below-average performers got the biggest boost from using AI tools.
Low performers + AI performed better than high performers without AI.
Training in AI improved both the quality and the speed of output for all consultants.
âWhen the consultants were asked to perform tasks outside of ChatGPTâs range of capabilities, the study found:
Those who used AI were more likely to trust ChatGPT even when the output was incorrect.
The AI was consistently wrong and also dangerously convincing and âconfidently correct.â
Training of the consultants in AI led to even greater levels of trust in the output even though it was consistently wrong.
There was less variety in the results that were more generic and similar across the groups.
â Hereâs The Thing: When you play to AIâs strengths, it can raise your performance and enable you to compete above your pay grade. It will also improve your productivity to do more with less.
On the flip side, the reverse is true. When you cross the line and take AI outside its comfort zone, youâre heading for trouble (see case of lawyer who used ChatGPT to prepare for a court case, didnât check it, used it, and is now in a heap of trouble.)
The issue is knowing where the line is.
This applies to all of the AIs, not just ChatGPT. Theyâre convincing and confident when theyâre wrong. Thatâs because they donât know theyâre wrong!
âSo, the key takeaway is: use AI to raise your game, but do so with a critical eye.
See AI as your multi-purpose Swiss Army productivity tool. It can do many things, but at the end of the day, itâs in your hands.
Thereâs no doubt that AI can both supercharge your output AND help you compete above your pay grade. Just be careful, thatâs all.
Read Harvard Business School Study
w/News
Global Powers Sign Declaration to Jointly Govern AI Risks at UK's Safety Summit
The United States, China, UK, EU and over 25 countries signed a declaration to collectively govern AI risks at the UKâs AI Safety Summit. Following on from President Bidenâs 100+ page Executive Order on Monday, this gathering of political and tech leaders marked a milestone in the debate on AI oversight and regulation. Notably, China was included in this first major West-led effort on AI safety, agreeing to increased collaboration. Known as the âBletchley Declarationâ, the agreement lays the groundwork for joint risk assessment and policies to ensure safe, ethical AI development. Whilst many commentators, including me, remain skeptical about geopolitical rivals cooperating in the AI arms race, thereâs no denying that was a significant moment/photo opportunity. Letâs see how long it lasts!
Sources: Reuters (Biden Exec Order)Â | AI Safety Summit | The King | Sunak Interviews Musk
Apple Is Playing The AI Long Game
Microsoft and Google have been the notable Big Tech front runners in the AI gold rush this year, closely followed by Meta. Last month Amazon joined the party and started to show its hand on AI, leaving a notable absentee - Apple. Thatâs not to say Apple arenât in the game. Every product Apple has is packed with AI. Just wait until SiriGPT hits every device! Itâs just that Apple donât classify or report on this tech capability as discretely âAIâ. Having said that, this week we got a glimpse into how Apple are playing the AI long game with the announcement of their new M3 AI supercharged chips powering the new MacBook Pro. Support for up to 128 gb of memory on these chips unlocks workflows previously not possible on a laptop. The Neural Engine is up to 60% faster than in the M1 family of chips, which is super helpful for image and video work. Apple Press Release
Breakthrough AI Learns Concepts from Words, Like Humans
A recent study from New York University revealed an AI model that mimics the way that a toddler learns a language. This is different to the conventional way that large language models learn. As I mentioned above, a toddler can learn and understand that a prairie dog is not a dog, but thatâs not easy for an AI. The model was trained to learned from mistakes in much the same way as humans learn a language when children. This breakthrough has significant implications for natural language processing and the development of AI to be more human-like.
Source: Singularity Hub
Sam Bankman-Fried Found Guilty Of Crypto Fraud
I know itâs not an AI story, but itâs a significant one in the world of crypto and Web3. This week a jury found Sam Bankman-Fried, the founder of one of the largest cryptocurrency exchanges called FTX, to be guilty of defrauding customers and investors. In a nutshell, SBF used customerâs crypto assets for risky investments, buying things and paying millions of dollars for celebrity endorsements. The former crypto executive faces up to 110 years in prison, although he wonât be sentenced until March next year.
Source: The Information
Five Snippets Of Tech News
Google has committed to investing $2 billion to AI startup and OpenAI rival Anthropic. Last month Amazon invested up to $4 billion in Anthropic.
Microsoft has announced their $30/month AI CoPilot for businesses. Microsoft 365 Copilot can summarise documents, generate emails, create plans from notes, and improve Excel analysis.
It started as a joke, but deepfake songs featuring Indian PM Modi have gone viral on the Indian internet in the run up to national elections. PM Narendra Modi is heard crooning in languages like Hindi, Tamil, and Telugu, thanks to AI-powered voice cloning tools.
Meanwhile, Scarlett Johansson is suing an AI app for cloning her voice in the latest non-consensual deepfake story.
Paul McCartney and Ringo Starr just released a new Beatles single featuring AI-generated John Lennon, completed using machine learning to isolate individual tracks from an unreleased demo. "Now and Then" stalled in the 1990s because analog tape technology couldn't separate Lennon's blended piano and vocals.
â 66% of interactions with customer service chatbots were rated 1 out of 5. Full research paper.
w/Productivity
Supercharged Productivity Tips and Tricks
Raycast
If youâre a Apple user then youâre probably using Raycast on your Mac instead of the native Spotlight, why wouldnât you be?? Raycast is free but the app has just added a paid layer that integrates with ChatGPT and GPT-4âs real-time web results. For $8/month you can now search the Internet from your Mac without opening a browser.-
Brave Browser
If privacy is your thing, then look at what Brave, the privacy-centric browser, have done with their new AI chatbot called Leo. It is designed to offer "anonymous and secure" assistance to its users, such as translating, answering questions, summarising web pages, and content generation. The unique selling point of Leo is that there is no recording of your conversations, nor is any of them used in AI training. Leo is based on Metaâs Llama 2 AI model and is free for all desktop users, but a premium version using Anthropicâs Claude Instant model (with faster responses) will be available for $15 per month. (NOTE: Iâve just ditched Microsoft Edge and moved back to Brave, Iâll let you know how I get on.)
Hal9
Are you an SME and want/need to get instant answers from your data but donât have your own data analysts to do it for you? Then meet Hal9, an AI tool that lets you chat with your enterprise databases using secure generative AI. Where Hal9 is different to using ChatGPTâs Code Interpreter for data analysis is that your data never leaves your database. Where as ChatGPT requires that you upload your commercially sensitive data to be analysed, Hal9 brings the power of AI to you, not the other way round.
FREE TO WISER! READERS: If youâre new to generative AI and want to know the basics of how to get started, what you can do with AI and sample prompts to get you goingâŚdownload your copy of The Beginnerâs Guide To ChatGPT for FREE!
đ Show Your Support For Wiser!
Thank you for reading Wiser! If you got value and would like to support what Iâm doing, do this:
Forward this email: send it to anyone you know whoâs interested in the tech economy.
Make a donation: go to BuyMeACoffee to make a donation in the form of a virtual cup of coffee, they only cost âŹ2 each. And who doesnât love a coffee, right?
Check out my website: Youâll find links to everything I do at rickhuckstep.com.