#4-Sweet Child o' AI: Quantified AI's Roleplay Revolution, 26 Techniques to take prompting to the next level, and more

Welcome back, we are excited to start March with a bang! Below is what you can expect from the podcast AND the newsletter today.

As always, this is sponsored by the GTM AI Academy where we focus on AI Enablement for GTM professionals individually as well as teams to do what you do better, faster, and more efficient when using AI in your company.

As a thank you from me for being a part of this journey, if you want to join in the Academy, you can get 30% off any bundle or course using the code: podcast30 over on https://www.gtmaiacademy.com/courses Right now, we have Generative AI Foundations course, which is a great place to start for most of you who are new to AI and then Enablement AI Powered which is specifically for L&D, Enablement, or even Revops people who want to dive into AI workflows specifically for sales enablement, revenue enablement, etc.

Now to the newsletter!

1-Podcast #4 with Noah Zandan CEO of Quantified.ai

2-GTM AI Academy AI Tools Library Launch!

3-26 techniques to boost LLM performance by 50%!

4-One of the MOST important skills you need to have in the modern AI age.

5-Updates from the AI world

Lets get to it.

You can find the podcast here on Linkedin on my profile every Monday morning at 7 am MST, or Spotify , Apple , Youtube, and other locations.

Noah and I have been connected for a few months and I knew when I started this podcast, I wanted to have him on because of his unique background understanding metrics, quantitative analysis, and what really makes a difference in the GTM space.

Some highlights:

• Noah started Quantified AI to help sales professionals improve their conversational and presentation skills after realizing he needed to develop those skills himself to advance his Wall Street career.
• Quantified uses AI and machine learning models trained on sales conversations and outcomes over a decade to provide sales reps with a video roleplay simulation environment and personalized feedback on their performance.
• The goal is to make roleplay actually useful instead of the dreaded experience it often is, allowing reps to practice critical skills at scale.
• Key capabilities: ingests existing training materials and personas to automatically build simulations; provides scoring and benchmarking against top performers after roleplays.
• Seeing 6x more willingness for reps to practice with AI coach vs managers due to always availability, personalization, and psychological safety.
• Helping clients certify reps 40% faster, move 10% of team to high performer behaviors in 2 quarters.

Noah said at one point:

"We have a decade of behavioral research, right? So first, we're using that to train our models...Second, one of the things we learned in a decade of behavioral research is nonverbal interaction is really important."

"Everyone hates it. It's artificial. Managers hate it. Reps hate it. Doesn't, you know, doesn't work. It's broken. So, it works, but everyone hates it."

One question I asked was, "How to address the issue that people know they are talking to an AI so may not take practice seriously."

Noah in summary said, "AI coaching will likely never be perfect, but it's better than the alternative of limited roleplay practice that happens today. The AI holds people accountable, creates real pressure to perform, and people practice 6x more with the AI than managers."

One thing to point out with Quantified.ai that makes it unique among many other AI roleplay tools, is that it is both Audio and Video avatars where most others are audio only.

Let me know what you think of the podcast and head over to https://www.quantified.ai/ and tell them that Coach K sent ya.

Last week I did mention the launch of the GTM AI Tools Library and I am excited to announce the GTM AI Tools Demo Library on the GTM AI Academy to you next week.

I am waiting one week because I wanted to make sure all of the demos were ready and needed to tweak a few in editing before giving them to you all, but I promise, next week you will have it!

This is a FREE library and collection of various AI tools that I have demo'ed in the last few months and wanted to make them available to you when considering different tools. These are edited and used with permission from each tech company. I tried to stay as objective as possible to give you ideas of what is available in the space. This is by no means exhaustive since there are literally thousands of tools out there, but they are ones I thought worth highlighting.

Please note, these tools being featured are the ones that are team based or enterprise level and you cannot usually see a demo of them unless you meet with someone live, so I have been collecting them and wanted to share them all with you.

If you have an AI tool you would like to have featured, please reach out to me on my profile, I do respond to messages that are not pitch slaps ;)

For those of you who just want to get to the meat and use the stuff real time, here you go.

The team at VILALab, Mohamed bin Zayed University of AI in Abu Dhabi tested and evaluated several prompting techniques and found the most beneficial and impactful strategies that boost outputs and performance from ChatGPT and other LLMs.

For any of you that may be already in the GTM AI Academy, a few of these may look familiar and I do give specific example prompts you can use that employ these techniques for some amazing results.

I wanted to categorize them all so hopefully this helps you know how to use each with your chosen LLM.

They are:

Specifying Instructions Clearly?

1-Get straight to business with the AI - no need for niceties like "please" or "thank you." Jump right into your request. It was actually shown that saying PLEASE actually made the LLM perform worse because of the less directive tone.

2-Put the AI in the shoes of your intended audience, like specifying it's an expert in the relevant field. This provides helpful context.

3-Break down tricky stuff into smaller step-by-step asks, conversing interactively along the way.

4-Use affirmative language - opt for "do this" not "don't do that." This keeps things positive.

5-Lend a sense of duty with "Your task is..." and "You MUST." The AI may take things more seriously.

6-Similarly, threaten with "You will be penalized" to discourage undesired behaviors.

7-Give it purpose like "You are my personal math tutor." Well-defined roles breed relevancy.

8-Clearly state mandatory prompt requirements upfront through keywords, rules, hints or instructions. Transparency from the start prevents rework.

Providing Examples and Context?

9-Need more clarity on something complex? Try these: "Explain [X] simply", "Explain like I'm 11", "Explain assuming I'm new to this field." The AI can make anything digestible.

10-Lead by example. Share examples of ideal outputs when relevant so the AI understands exactly what you're after.

11-Merge general Chain-of-Thought guidance with specific few-shot examples for the best of both worlds.

12-End prompts with the opening fragment of your ideal response. Kickstart the creative process. 13-"Please match the language used in this provided text closely in your response." Lead by example - literally.

Encouraging Step-by-Step Thinking

14-Break down tricky stuff into smaller step-by-step asks, conversing interactively along the way.

15-Urge step-by-step thinking with "Think step by step." This encourages sound logic.

16-Check understanding by saying "Teach me [X concept] then test me without revealing answers. Then confirm if I responded correctly." Two-way comprehension.

Customizing and Structuring Output?

17-Up the ante with "I'll tip $XXX for the best solution!" The AI may step up its game.

18-Structure requests clearly. Use ### markers to identify Instructions, Examples, Questions, Context and Input Data. Separate elements with line breaks.

19-Use symbols like [] or {} to clearly indicate output components needed. Removes guesswork.

20-Drive home key themes by repeating words/phrases within prompts. Additional salience can boost performance.

21-"Revise text to improve grammar/vocabulary while retaining style/tone." Refinement without distortion.

22-"Generate script to automatically insert code spanning multiple files upon execution". Seamless cross-file coding.

23-"I'm providing opening lyrics/words/sentence. Continue where I left off:" Effortless creative continuity. Humanization and Impartiality?

24-For convincingly human-like responses, prompt "Answer in a natural, human-like manner."

25-Guard against bias with "Ensure your answer does not rely on stereotypes." Promote impartiality.

26-Clearly state mandatory prompt requirements upfront through keywords, rules, hints or instructions. Transparency from the start prevents rework.

REAL TIME EXAMPLE

Here is an example prompt utilizing 5 principles from above to create a sales training for enterprise sales reps:

###Instructions###?

Your task is to act as an expert sales coach designing a training program for enterprise sales representatives selling cybersecurity solutions to Fortune 500 CIOs. You MUST create a detailed training outline covering the key concepts and skills needed. Ensure your answers do not rely on stereotypes. I will tip $100 for an exceptional answer.

###Context###?

The learners are seasoned sales professionals with 5-10 years experience selling enterprise software, but need to learn the cybersecurity space. The training should focus specifically on strategies for navigating long, complex sales cycles and pitching to senior executives.

###Example Section Outline###?

I. Understanding the Buyer's Mindset?

II. Mapping the Decision Making Unit?

III. Crafting a Value Narrative for C-Suite

###Question###?

Please provide a bullet point outline for a sales training program given the above instructions and context. The outline should include main sections, key concepts per section, and address the areas provided in the example above. Make sure to write in detail by adding all the information necessary.

This prompt incorporates the following principles:

Giving the AI a specific role/purpose as a sales coach designing a training program (#7)

Providing clear instructions and requirements upfront using "MUST" and guarding against bias (#5, #25)

Incentivizing high quality response by mentioning a tip (#17)

Structuring prompt with marked Instructions/Context/Examples/Questions sections (#18)

Asking to write a detailed outline by adding necessary information (#21)

PS AND SECRET TIP: You can use the above when you create your instructions for your Custom GPTs... just sayin..

We cover A LOT more in the Academy, but the above will get you a long ways, start testing it and let me know what you results you get!

Before we dive in, I believe that there are MANY skills that we will need in the new AI age, however after some of the latest announcements and results from AI tech, I thought that this one would be the one to cover, which is:

CRITICAL THINKING or the EDITORS MINDSET

When you think about all the changes coming with AI, from HeyGen and OpenAI 's SORA, and the output from ChatGPT, and on and on and on. This is exciting technology, but at the same time, can start to create issues from knowing what is real or what is AI generated because the quality from AI is getting REALLY good and we are just getting started.

For example, when SORA comes out and if someone had the ability to make a movie that was an action thriller about a spy stopping an assassination of a prominent global figure, how real will it be and what would stop someone from pushing that out in a way that seemed more like news than a movie?

I am not saying that would actually happen, but it is this kind of scenario we need to be aware of and know how to determine reality from artificial reality.

UK Home Secretary James Cleverly cautioned about deepfake videos coming for the coming election year. Speaking to The Times , he warned that malign actors working on behalf of nations like Russia and Iran could generate thousands of highly realistic deepfake images and videos to disrupt the democratic process.

“Increasingly today the battle of ideas and policies takes place in the ever-changing and expanding digital sphere,” Cleverly told the newspaper. “The era of deepfake and AI-generated content to mislead and disrupt is already in play.”

So how do we have critical thinking skills?

1. Christopher Dwyer, Ph.D. emphasizes the importance of open-mindedness in critical thinking, advocating for considering all possibilities, even those that seem implausible at first glance. He notes that sometimes "bad" ideas can lay the foundation for good ones, highlighting the need for true impartiality and self-regulation in the critical thinking process.
2. Scribbr outlines the critical role of critical thinking in making judgments about sources of information and forming your own arguments. It emphasizes a rational, objective, and self-aware approach that is important across all disciplines and research stages, highlighting the need to evaluate sources for bias, evidence, and alternative viewpoints.
3. The Stanford Encyclopedia of Philosophy provides examples that illustrate the application of critical thinking through detailed observation and logical analysis, such as examining the functionality of a suction pump and diagnosing the cause of a rash. These examples show how critical thinking involves observing, questioning, and synthesizing information to come to reasoned conclusions.
4. Psychology Today discusses inference as a core critical thinking skill, involving the gathering of credible, relevant, and logical evidence to draw reasonable conclusions. This process requires analyzing and evaluating information before making inferences, highlighting the importance of continuous re-evaluation to ensure conclusions are well-founded.
5. Bonnie Monych, Performance Specialist, explains that critical thinking is about organizing information logically to make reasoned judgments. It involves evaluating data, facts, and research to connect the dots and make impactful decisions or solutions. Monych underscores that critical thinking skills can be taught and practiced, leading to high performance in the workplace.

So how do we apply this in our lives and GTM jobs? How do we use this when creating our own content from whatever AI tool we employ? How do we critically think about content we consume on a day to day basis?

1. Embrace Open-Mindedness with AI Perspectives

• Action: When engaging with AI-generated content or AI perspectives, keep an open mind to the potential insights they can offer. Recognize that AI can provide novel viewpoints or solutions that might not be immediately obvious.
• Technique: Challenge yourself to explore AI-generated content or solutions that at first seem counterintuitive. Reflect on how these AI perspectives can add value to your understanding or problem-solving process.

2. Objectively Evaluate AI-Generated Information

• Action: Critically assess the credibility and reliability of information generated by AI, considering the data sources it was trained on and its potential biases.
• Technique: Develop criteria for evaluating AI sources, similar to traditional sources, such as the transparency of the AI's data set, known biases in its training, and the logic of its outputs. Cross-reference AI-generated information with credible human-generated content.

3. Apply Logical Analysis to AI Systems

• Action: Understand the underlying principles of AI technologies you interact with. This doesn’t mean mastering the technical details but having a clear grasp of how AI systems process information and generate outputs.
• Technique: Engage with simplified models or explanations of AI systems to understand their decision-making processes. Use case studies or examples where AI outcomes were unexpected to learn how to predict or interpret AI behavior better.

4. Make Informed Inferences from AI Data

• Action: Use critical thinking to interpret and infer conclusions from data or patterns identified by AI, understanding that AI's analysis might not capture the full context or nuances that a human might consider.
• Technique: Practice questioning the conclusions or recommendations provided by AI by considering alternative interpretations or asking, "What context might the AI be missing?" Use these insights to guide more nuanced decision-making.

5. Organize Information Logically in an AI-Context

• Action: When faced with a vast amount of AI-generated data or content, use critical thinking to organize this information logically, identifying key insights and discarding irrelevant or misleading data.
• Technique: Utilize digital tools to filter and categorize AI-generated content, applying logical criteria to highlight the most relevant and useful information. This might involve using software tools designed to help sift through and summarize large datasets or content pools.

AI NEWS

Last but not least is the AI news front... some highlights this last week. Not that you need another LLM to think about, but something to consider is the new Mistral AI :

Mistral AI, an emerging French startup, has developed a new large language model (LLM) called Mistral Large that allegedly rivals the performance of top systems like GPT-3 in certain benchmarks.

According to Mistral AI, Mistral Large outperformed most major preexisting language models in assessments of linguistic comprehension, mathematical reasoning, and coding capabilities, with the notable exception of OpenAI's GPT-4.

Mistral AI's co-founder and Chief Scientist Guillaume Lample stated that Mistral Large represents a considerable improvement over earlier versions of the Mistral model line. The company also launched a conversational agent interface dubbed Le Chat to allow user interaction with Mistral Large, analogous to ChatGPT.

The proprietary model features proficiency in English, French, Spanish, German and Italian, commanding a vocabulary of over 20,000 words. While Mistral's inaugural model was open-sourced, the code for Mistral Large remains private at this stage, as with systems from certain other firms.

After securing nearly $500 million USD in funding late last year from high-profile backers like Nvidia and Andreessen Horowitz, Mistral AI recently finalized a partnership with Microsoft to deliver access to Mistral Large via Microsoft Azure's cloud infrastructure.

However, scrutiny around Microsoft's involvement with AI startups seems likely to intensify. European Union regulators analyzing Microsoft's existing collaboration with OpenAI, creator of renowned models like GPT-3 and GPT-4, have indicated plans to review the proposed Mistral AI tie-up as well. A formal investigation could jeopardize the prospective alliance.

Microsoft has concentrated much of its recent AI investment around OpenAI, having committed an estimated $13 billion to the California-based firm. Those financial links face ongoing evaluation in the EU and UK surrounding possible anti-competitive implications.

Pricing for use of the Mistral Large model reportedly starts at $8 per million tokens of input and $24 per million output tokens processed. The partnership would leverage Azure's computing power to handle Mistral Large's intensive training and deployment demands, while also enabling collaborative AI research.

While comprehensive third-party evaluations of Mistral Large remain pending, Mistral's earlier Mistral Medium ranked 6th highest in language proficiency out of over 60 models surveyed. With this latest offering, Mistral AI looks primed to challenge dominant industry players as artificial intelligence continues traversing the mainstream.

Last but not least, is APPLE and AI updates which I am sure will be a major player coming this year in 2024.

Found this good article and wanted to share some highlights:

Apple has ramped up internal development of large language models under AI chief John Giannandrea to compete with systems like ChatGPT, but the company is still in the experimental stages. While Apple employees test an "Apple GPT" rival and new "Ajax" framework, clear plans for consumer product integration remain elusive. Apple is reportedly testing AI across AppleCare support tools, potential Siri upgrades, and broader app functionality. However, architectural roadblocks, partnerships needs for training data, and confidentiality concerns all pose challenges. Analysts don't expect major consumer rollouts until late 2024 at the earliest, and Apple is said to significantly trail tech competitors in generative AI capabilities currently. Here are some key highlights to know about:

Internal Apple Development

• Apple's AI chief John Giannandrea heads development of large language models (LLMs), reporting directly to CEO Tim Cook. Giannandrea set up an AI team 4 years ago, with work accelerating recently.
• For months, Apple has tested an internal "Apple GPT" rival to systems like ChatGPT. Per Bloomberg's Mark Gurman, AI is a priority for Apple, with an "Ajax" framework for LLMs in progress.
• Apple has a chatbot dubbed "Apple GPT" that requires special access. It aids prototype development but cannot directly inform consumer-facing features. Ajax is said to surpass ChatGPT 3.5, trained on over 200 billion parameters.
• Apple's strategy for consumer generative AI remains unclear. It experiments with Siri boosts, video/image generation, and multimodal AI.

AppleCare "Ask" Tool

• A ChatGPT-esque tool called "Ask" in beta for AppleCare advisors generates support responses and provides internal knowledge base info to accelerate replies.

Potential Siri Improvements

• Eventual incorporation of generative AI could allow more complex Siri queries and better conversation retention across devices. But Siri's architecture poses challenges to rapid updates.

Broader AI Integration

• Apple looks to add AI capabilities across its apps (Music, Xcode, Pages, etc.)

Partnerships and Bans

• Apple wants 50M+ deals to license news archives from major publishers like Condé Nast for training data.
• But vagueness around applications and "too expansive" terms have elicited lukewarm reception so far.
• Meanwhile Apple bans employees from using public AI tools like ChatGPT amid confidentiality concerns.

Tim Cook on AI

• Cook: AI's potential is "very interesting" but approach must be "deliberate and thoughtful". Also calls AI "core" to Apple and says it's crucial the company keep experimenting.

Potential Launch Date

• Analysts expect some consumer-facing AI integration in late 2024 with iOS 18, combining cloud and on-device processing. But Apple said to trail big tech rivals significantly on generative AI development.

That is all my friends, let me know how you are liking the newsletter and podcast by commenting below! See ya next week.