5 | Beatles last song? | The AI Dilemma |

This week, The 🎶 Beatles 🎶 made a surprising move by releasing a new song featuring the full band – yes, you read that right! (Happened Yesterday) 

Leaders from developed nations gathered to discuss the effects of AI on our society. In light of this, we've shared a notable TED Talk on the <AI Dilemma>. Although its narrative somewhat contradicts the democratization of LLMs—a stance I, as a writer, do not align with—I've shared it to foster transparency, allowing you to draw your own conclusions

On November 6th, OpenAI is slated to unveil a major 'Killer' feature for ChatGPT. For those keen on brushing up on the basics of using ChatGPT, our actionable tip of the week is sure to brighten your day.

Our Menu :)

• Weekly Digest:

  • The AI Dilemma 🤖.

  • British PM Rishi Sunak secures ‘landmark’ deal on AI testing 🌎

  • Beatles' Last Song 'Now and Then': A Timeless Melody Revived by AI. 🎶

• Actionable Tips:

  • Powerful ChatGPT Prompting in 5 Minutes, Don’t Buy Them! 💡

• Training & Crash Courses:

  • LLM University / Huggin Face 🏫🤗

WEEKLY DIGEST
The AI Dilemma: Navigating the Perils and Promises of Advanced Technology

At the CogX Festival 2023, Tristan Harris, a prominent advocate from the Center for Humane Technology, delivered a compelling address on the challenges posed by AI. The talk, centered on the urgent need for a public discourse and strategic action in the face of rapid AI advancements that threaten to outpace our societal, ethical, and regulatory frameworks.

Why Does It Matter?

A staggering revelation from a survey conducted by AI Impacts in August 2022, indicated that half of the AI researchers believed there's a considerable risk of human extinction or severe disempowerment due to our potential inability to control AI.

Social Media: The First Contact with AI

Social media platforms have served as the initial interface between AI and humanity. The promise was alluring: to amplify our voices, connect with our friends, and join like-minded communities. For small and medium-sized businesses (SMBs), it was the golden ticket to the global stage, enabling them to connect with customers beyond the confines of geography.

Yet, as these platforms began to harness AI to captivate us further, they inadvertently sparked a relentless race for attention. The algorithms, engineered to prioritize engagement, learned to feed us an endless stream of content calibrated to our interests and biases. It was a spectacle that knew our cravings better than we did ourselves, often serving as a diet of digital junk food over substantial nourishment.

This insatiable Race For Engagement ignited a competition that exploits our primal instincts rather than celebrating our higher aspirations. The result? A tangled web of content that lures us in, not to enlighten, but to ensnare our time and attention

This race for engagement not only produced these effects but also captured the core functions of our society, now:

• Elections are run through this engagement economy

• Children's development happens through the engagement economy

• Democratic discord happens through the engagement economy

• …

Creation Economy with LLMs: The Second Contact with AI

After social media introduced us to the power of AI, the advent of Large Language Models (LLMs) like GPT-3, and subsequently GPT-4, marked a new era in our interaction with artificial intelligence.

We do it because the promises are beyond imagination but start with:

• AI will make us more efficient

• AI will help us code faster

• AI will find cures for cancer

• AI will solve climate change

• Ai will increase the GDP

In boardrooms and startups alike, LLMs are celebrated for transforming creative processes and automating complex tasks, thereby leveling the playing field between small businesses and industry titans. These tools are not only augmenting human potential but also reshaping the very nature of innovation and productivity.

Race To Deploy

Once more, the frenetic pace at which these AI capabilities are being rolled out comes with significant caveats. The rapid integration of such powerful technologies has outstripped the development of corresponding ethical frameworks and safety mechanisms.

There is a tangible risk of exacerbating societal inequities, entrenching biases, and fostering an environment rife with misinformation—consequences that could undermine the initial promise of AI.

Emergent Capabilities and the Theory of Mind

In a nutshell, the concept of theory of mind is the ability to understand and model what others are thinking or feeling.

The progression is striking: in 2019, AI resembled a one-year-old's cognitive empathy, barely scratching the surface of understanding. By 2020, it advanced to a four-year-old's level, and further to that of a near seven-year-old in 2022. The leap was even more profound with the advent of ChatGPT in November 2022, which showcased the mental modeling capabilities akin to a nine-year-old.

However, the real breakthrough came with GPT-4. Post its release, AI's theory of mind was assessed to exceed that of an average adult. This milestone suggests that modern AI can not only infer what others are thinking but also engage in sophisticated, strategic thinking. Harris's observations underscore a monumental shift in AI cognitive abilities, marking a new era where AI's understanding of human psychology is profoundly nuanced.

The Dangers of Rapid Deployment

The rapid adoption of AI technologies has far-reaching implications. ChatGPT, for instance, reached 100 million users in a mere two months, surpassing the user acquisition rates of platforms like Netflix and Facebook. AI's integration into consumer products, such as Snapchat incorporating GPT-4, signifies the speed and scale at which these technologies are entering everyday life.

The same principle applies to open-source LLMs, such as the one Facebook has invested approximately $35 million in training the model. Named LLAMA2, this model incorporates robust safety protocols. Yet, as Harris pointed out, a mere one hundred dollars and a few days were sufficient for an engineer to subvert these protocols, creating a variant dubbed 'BAD-LLAMA.' This iteration was capable of instructing on how to synthesize anthrax, highlighting the potential dangers of even well-guarded AI systems.

Even the engineers that are building this AI, are not able to predict which possibilities are going to pop out. In other words, as they train GPT4.5 or GPT5, they do not know whether these new AIs will have a set of dangerous capabilities.

The Path Forward

• Coordinate the Race: Harris's insights draw a parallel with the 'Social Dilemma,' highlighting the inherent problems in social media companies' business models and their pervasive effects. It's imperative to avoid a similar trajectory with AI. We must foster real commitments among all stakeholders, including tech companies, governments, and civil societies worldwide. By collaborating, we can navigate the competitive landscape responsibly, ensuring AI advancements serve the common good rather than exacerbate global inequities or conflicts.

• Emergency Brakes: Progressing through GPT-5, GPT-6, and beyond, we are bound to witness the emergence of unforeseen capabilities. It is crucial to install systemic 'emergency brakes' to halt or redirect developments that threaten to spiral out of control. This includes rigorous monitoring systems, regulatory frameworks, and the willingness to pause when necessary to assess and address potential risks.

• Limits on Open-Source: While open-source has been a bedrock for innovation, allowing for democratization and rapid advancement in AI, it also poses significant risks. Harris emphasizes the need for sensible boundaries. The accessibility of powerful AI tools must be balanced with safeguards that prevent misuse. This could mean restricted access to certain features or capabilities that have higher risks associated with them.

• Liability: As AI systems become more integrated into our lives, assigning liability for their actions becomes increasingly complex. Harris suggests a legal framework that can keep pace with AI's evolution, ensuring that accountability is not lost in the complexity of these systems. Developers, deployers, and users must all bear responsibility for the outcomes of AI, with clear guidelines on how liability is determined and enforced.

• Mitigation Strategies: Finally, Harris calls for proactive mitigation strategies. We must anticipate challenges and design AI systems with the resilience to withstand and adapt to adversities. This entails investing in AI education, fostering interdisciplinary research on AI impacts, and developing best practices for AI ethics and governance.

In conclusion, 'The AI Dilemma' requires us to be as innovative in our management and oversight of AI as we are in its creation. Harris's message is a potent reminder that our future with AI is not predestined—it is a product of the choices we make now. By embracing cooperative strategies, building robust safety nets, and upholding our shared human values, we can ensure that AI becomes a force for good in the symphony of human progress."

WEEKLY DIGEST
British PM Rishi Sunak secures ‘landmark’ deal on AI testing

The Bletchley Declaration emerged from the 2023 AI Safety Summit, marking a pivotal moment in global AI governance. This accord, endorsed by 28 nations including major players like the US, China, and the EU, sets the stage for collaborative efforts in AI safety and regulation.

What’s New:

1. US Influence: The US showcased its leading role in AI, with VP Kamala Harris indicating US domestic AI policies could serve as a global model.

2. UK's Innovation-Focused Approach: The UK emphasized its innovation-first approach, aiming to boost its AI businesses before passing new AI laws.

3. Open Source AI Development: France and some nations advocated for open source AI development to ensure transparency and competitive equity in AI technologies. France’s AI darling, Paris-based Mistral, is an open-source startup.

4. Protectionism: The UK and the US announced plans for national AI safety institutes to address societal risks and national security concerns related to AI​

Why Does It Matter?

The Bletchley Declaration signifies a proactive approach towards establishing a common ground for addressing AI safety concerns on a global scale. This collaborative endeavor underpins the critical importance of aligning international efforts to ensure that the deployment of AI technologies is carried out safely and responsibly

WEEKLY DIGEST
Beatles' Last Song 'Now and Then': A Timeless Melody Revived by AI

Nov 2nd, 2023. Paul McCartney and Ringo Starr have released a new Beatles single titled "Now and Then," featuring the magical touch of AI on John Lennon's vocals, which were extracted from an unreleased demo from the 1970s.

What’s New: 

• The song "Now and Then" traces back to a demo recorded by John Lennon in the '70s. It remained incomplete until recent advancements in AI audio technology facilitated the extraction of Lennon's vocals using AI-powered stem separation technology.

• Lennon's vocals were obtained from a demo cassette handed over to Paul McCartney by Yoko Ono, Lennon's wife, in 1994. McCartney and Starr, with a little help from AI, breathed new life into the old demo, adding new bass, drums, guitar, and strings, while also incorporating a recording of George Harrison's electric and acoustic guitar from 1995. The recording sessions took place at Capitol Studios in Los Angeles​.

• Released on November 2, 2023, as a double A-side single alongside a new mix of "Love Me Do," the Beatles' debut single from 1962, "Now and Then" is being hailed as the last Beatles song. It will also feature in the expanded re-issue of the 1973 compilation 1967–1970.

Why Does It Matter?

Sixty years since the dawn of Beatlemania, the melody unveils the splendid tapestry of AI, weaving lost memories into new tales, and orchestrating unison even when the maestros have ventured beyond the stars.

The blend of AI and harmonies may dance upon the edges of tradition, yet the celestial reunion of the Beatles in "Now and Then" unveils a heartwarming ode to the endless dance between the past and the present. It's a poetic testament to how the quill of technology can script timeless tales, rekindling the embers of bygone eras in the canvas of today.

And as the chords of 'Now and Then' echo through the corridors of time, music aficionados and dreamers are invited to a grand ball, to waltz with the timeless resonance of the Beatles' harmony, and muse upon the uncharted symphonies that AI could compose in the concert of the morrow.

Listen:

ACTIONABLE TIP
Master ChatGPT Prompting in 5 Minutes, Don’t Buy Them!

Here's a simple breakdown:

1. Identify Your Goal:

   • What do you want ChatGPT to do? Generate content, answer questions, create plans?

2. Be Clear and Precise:

   • A vague request gets a vague answer. Be clear about what you want.

   • Example: Instead of saying "Write about dogs", say "Write a 500-word article about the benefits of adopting rescue dogs".

3. Guide the AI:

   • Use guiding words to shape the response.

   • Example: To get a creative story, you might start with "Once upon a time in a city full of whimsical creatures…"

4. Customize Your Prompts:

   • Add placeholders where you want personalized input.

   • Example: "Write a poem about [your topic] that is [number of lines] lines long."

5. Practice and Experiment:

   • The more you practice, the better you’ll get at crafting prompts that get you the results you want.

SEO Article Generation

Act as an expert SEO content writer who creates optimized articles. Write a 800 word article in [language wanted] about [topic] that will attract potential clients to my [industry] website. Make sure to include relevant keywords and optimize the article for searches related to [keyword]. End the article with a call to action for users to subscribe to my newsletter for more valuable information.

Unlimited Prompt Generation

Act as an expert prompt engineer with deep knowledge of ChatGPT capabilities. Create a table with two columns - one for prompt function, and one for the actual prompt. Fill it with 30 rows of prompts focused on [social media content creation]. Make each prompt detailed and useful for generating the desired content. Put bracketed placeholders for customization.

Content Exam Generation

Analyze the content from this transcript. Generate [10] multiple choice exam questions based on the key concepts, with 3 possible answers each. After each question, wait for my response, indicate if it was correct or incorrect, and if incorrect, ask me to try again while explaining why my answer was wrong. At the end, provide an overall score for my exam performance

Social Media Plan Creation

Act as a social media expert and create a one week social media plan for [Instagram] for [small business], with the goal of [goal]. Include posts, stories and reels, deciding yourself the optimal frequency and volume. For each post, provide the text, hashtags, and a prompt description for the visual that could be used with AI image generation. Deliver the plan in a table format.

Virtual Personal Trainer

I am [context about you]. My fitness goal is [goal]. Create a one week fitness plan for me, considering I can dedicate [time per week] to exercise. Account for [limitations]. Provide the plan in a table format with day, exercises, sets/reps, and links from PlanFit for guidance. Also create a menu with breakfast, lunch, and dinner to help me meet my goals. Review the PlanFit recommendations and make any changes needed to optimize the plan.

TRAINING
LLM University / Huggin Face 🎉

Large Language Models (LLMs) have recently become a hot topic and a popular technological solution in conversations. To help you delve deeper into LLMs, we’ve curated three free courses. For those keen on mastering prompt engineering, we have an additional practical list.

1. Cohere LLM University

   You'll learn:

   • How do LLMs work?

   • What are LLMs useful for?

   • How to use LLMs to build and deploy apps?

1. Hugging Face NLP course

   This course covers:

   • 🤗 Transformers

   • 🤗 Datasets

   • 🤗 Tokenizers

   • 🤗 Accelerate

Thank you, see you next week!