The Rise of AI: SaaStock Local

Slide 1

Slide 1 text

Bastian Grimm | Peak Ace AG | @basgr The Rise of AI Tools, Strategies & Tips to Drive Growth June 21st 2023

Slide 2

Slide 2 text

A bit of a warmup…

Slide 3

Slide 3 text

Stock Photo or AI? Which image was created by a machine/AI?

Slide 4

Slide 4 text

VS #A #B

Slide 5

Slide 5 text

VS #A #B

Slide 6

Slide 6 text

VS #A #B

Slide 7

Slide 7 text

VS #A #B

Slide 8

Slide 8 text

pa.ag @peakaceag 8 Midjourney V5 has got REALLY good… just wow! Our all-time favourite AI text-to-image generation tool that's available right now: Source: https://www.midjourney.com/ & https://pa.ag/3C11Zwo

Slide 9

Slide 9 text

pa.ag @peakaceag 9 Back when it all started with DALL-E in April '22 DALL-E was primarily responsible for the fact that "mainstream media" started talking lots more about AI as a topic in general: Source: https://pa.ag/3DTj2RR

Slide 10

Slide 10 text

pa.ag @peakaceag 10 DALL-E & Co are/will be deeply integrated to MS products Microsoft recognised the potential of OpenAI early on, as they already invested in it back in 2019. In January 2023, MS invested another $10 billion – just because… Sources: https://pa.ag/3J6pSGb & https://pa.ag/3ZDKJaT Both Microsoft Designer and Image Creator are powered by DALL-E 2 - the AI art generator made by OpenAI. Microsoft invested $1 billion in OpenAI in 2019 and has an exclusive license to use.”

Slide 11

Slide 11 text

pa.ag @peakaceag 11 Actually, it‘s much more than that: meet MS365 Copilot This isn't strictly a topic for marketing, but "interacting with AI“ will soon be part of your daily (work) activities – and you might not even know it: Sources: https://pa.ag/3ASowLb & https://pa.ag/43z5tC0 Invoking Windows Copilot is easy – the button is front and center on your taskbar – simple to find and use. Once open, the Windows Copilot side bar stays consistent across your apps, programs and windows, always available to act as your personal assistant.

Slide 12

Slide 12 text

pa.ag @peakaceag 12 Various alternatives, e.g., Dream Studio by stability.ai Source: https://pa.ag/3fybYlm

Slide 13

Slide 13 text

pa.ag @peakaceag 13 Need help and/or inspiration? Try DS Prompt Builder: Source: https://pa.ag/3JbNxFM Super-simple drag & drop prompt building based on visual elements to choose style, lightning, etc.

Slide 14

Slide 14 text

pa.ag @peakaceag 14 Creating AI prompts is a key skill for any generative AI Whoever formulates the best, most creative and most unique AI prompts will create the best content (and save AI/API credits) Source: https://promptbase.com There is a marketplace (promptbase.com), which is dedicated to trading prompts – use it for inspiration!

Slide 15

Slide 15 text

Each AI has its strengths and weaknesses… The challenge: What to use when?

Slide 16

Slide 16 text

pa.ag @peakaceag 16 BlueWillow.ai to combine the power of multiple models such as DALL-E, Stable Diffusion and more Source: https://www.bluewillow.ai

Slide 17

Slide 17 text

Ok… great. What about real-world use cases though?

Slide 18

Slide 18 text

pa.ag @peakaceag 18 If you need product ads/images fast, try flair.ai AI-generated product photography using a drag-and-drop approach that allows you to place product imagery in literally any environment you can think of: Source: https://pa.ag/3ZNrFqi

Slide 19

Slide 19 text

pa.ag @peakaceag 19 cleanup.pictures Remove any unwanted object, defect or even people from your picture in seconds Source: https://cleanup.pictures

Slide 20

Slide 20 text

pa.ag @peakaceag 20 Another very practical use-case: Scan me, I am a very beautiful QR Code! Source: https://pa.ag/3Xjna6Z

Slide 21

Slide 21 text

General use-cases include building header images (for blog posts), email & social images, and quick image modifications for your marketing efforts What (else) to use it for?

Slide 22

Slide 22 text

pa.ag @peakaceag 22 Interesting but equally scary use-case from Osaka Univ. “We propose a new method based on a diffusion model (DM) to reconstruct images from human brain activity obtained via functional magnetic resonance imaging (fMRI)” Source: https://pa.ag/3ZBaHvG Reconstructing visual experiences from human brain activity offers a unique way to understand how the brain represents the world, and to interpret the connection between computer vision models and our visual system.

Slide 23

Slide 23 text

Speaking of scary stuff…

Slide 24

Slide 24 text

50% of AI researchers believe there’s a 10% or greater chance that humans will go extinct from our inability to control AI. Source: https://pa.ag/41VW8DV

Slide 25

Slide 25 text

A bit like… catching a flight when 50% of engineers think there's a 10% chance that everyone on it will die. Would you get on the plane? Source: https://pa.ag/42hDGWm

Slide 26

Slide 26 text

Exaggeration? Maybe…

Slide 27

Slide 27 text

pa.ag @peakaceag 27 Voice- & face cloning is already very much possible Granted, it’s not perfect just yet, but it’ll do the trick. Solutions such as ElevenLabs.io or frameworks such as DeepFaceLab are available for free/at-minimum cost, to everybody: Source: https://pa.ag/3mIlAgF Good afternoon, everybody, this is President Obama speaking! My dear friend Bastian asked me to say hello. De-age the face Replace the head

Slide 28

Slide 28 text

pa.ag @peakaceag 28 AI-powered face-swapping tech caused $622k fraud In China, AI-powered face-swapping was used to impersonate a friend of a victim during a live video call and receive a transfer of 4.3 million Yuan: Source: https://pa.ag/3WH8wpz

Slide 29

Slide 29 text

pa.ag @peakaceag 29 AI-generated image caused a brief stock market dip A fake image showing an explosion near the Pentagon was shared by multiple verified Twitter accounts last Monday, causing confusion and a brief dip in the stock market: Source: https://pa.ag/434n6db

Slide 30

Slide 30 text

So, just imagine what’s out there that we don‘t have access to…

Slide 31

Slide 31 text

pa.ag @peakaceag 31 What did I bring for you today? A bit of theory Must-have tools Real-world use cases Future outlook AI vs. Search

Slide 32

Slide 32 text

Some more theory first

Slide 33

Slide 33 text

pa.ag @peakaceag 33 A timeline of AI and language models Source: https://pa.ag/3Ld7yxv

Slide 34

Slide 34 text

pa.ag @peakaceag 34 Larger data sets = more innovations in NLP For example, MUM uses a text-to-text Transfer Transformer (T5) trained on the Colossal Clean Crawled Corpus (C4), which is significantly larger than BERT's training data Source: https://pa.ag/3EFfRfT (MUM) Basis: Web crawl (220 TB) with approx. 3 billion web pages (BERT) Basis: Wikipedia with approx. 56 million articles VS From a marketing standpoint it's easy to say that MUM is 10 times smarter than BERT…

Slide 35

Slide 35 text

Megatron-Turing NLG (Microsoft & Nvidia): 530 billion Pathways Language Model 2 (Google): 540 billion WuDao 2.0 (Beijing Academy of AI): 1.5 trillion There are many more...

Slide 36

Slide 36 text

pa.ag @peakaceag 36 But: OpenAI estimates diminishing returns on scaling size “I think we’re at the end of the era where it’s going to be these, like, giant, giant models,” Altman told an audience at MIT, “We’ll make them better in other ways.” Source: https://pa.ag/42dtttO

Slide 37

Slide 37 text

Newly built from the ground up with true multimodality, high efficiency in using external tools/APIs, memory capabilities, etc. Google is working on a brand- new model: Gemini

Slide 38

Slide 38 text

pa.ag @peakaceag 38 Quality vs. quantity: Breaking down training data sources “A Comprehensive Analysis of Datasets Used to Train GPT-1, GPT-2, GPT-3, GPT-NeoX- 20B, Megatron-11B, MT-NLG, […]“ Source: https://pa.ag/3UId4KA

Slide 39

Slide 39 text

pa.ag @peakaceag 39 Can you really trust the info in those training data sets? The idea: to precisely time malicious modifications just before a snapshot for inclusion in a web-scale dataset, aka “Frontrunning data poisoning”: Source: https://pa.ag/3LuJJ4L If an attacker can precisely time malicious modifications just prior to a snapshot for inclusion […], they can front-run the collection procedure. […] Even if a content moderator detects and reverts malicious modifications after-the-fact, the attacker’s malicious content will persist […]

Slide 40

Slide 40 text

pa.ag @peakaceag 40 Problems & challenges with (large) language models ”Training a single BERT base model (without parameter tuning) on GPUs was estimated to require as much energy as a trans-American flight.” [Source: Mozilla] Source: https://pa.ag/3tuOwJ9 Computing power The quality of the results improves as the size of the LLMs increases. Gains are incremental, while costs increase exponentially. Biased training data If biased data is used, the system will likely show the same bias when making decisions in practice. Environmental impact Training a large transformer model produces 5,680% more carbon emissions than one human per year. Static perspectives Large language models are usually trained once and used over long periods of time.

Slide 41

Slide 41 text

pa.ag @peakaceag 41 A more hands-on “bias example” from Amazon The company developed an AI to help scale recruiting – which didn’t work so well… Source: https://pa.ag/4228GtP Amazon’s computer models were trained to vet applicants by observing patterns in resumes submitted to the company over a 10-year period. Most came from men, a reflection of male dominance across the tech industry. […] In effect, Amazon’s system taught itself that male candidates were preferable. It penalised resumes that included the word “women’s” and downgraded graduates of two all-women’s colleges.

Slide 42

Slide 42 text

Have I scared you enough to not blindly trust anything AI by now?

Slide 43

Slide 43 text

Next up: Tools

Slide 44

Slide 44 text

pa.ag @peakaceag 44 Source: https://pa.ag/3yrqQYn theresanaiforthat.com – yup… there is! If you prefer searching, this might be of help:

Slide 45

Slide 45 text

Be very, very mindful of the data you put into any of those tools. #confidentiality #dataprotection

Slide 46

Slide 46 text

AI for: Text copy

Slide 47

Slide 47 text

pa.ag @peakaceag 47 Overview of AI-based content generation tools Depending on your needs and desired outcome, I am sure you’ll find one you like:

Slide 48

Slide 48 text

Use-cases for Jasper, for example, include helping you write sales copy for outreach emails or content (drafts) for a blog post, etc. What to use it for?

Slide 49

Slide 49 text

Never publish without checking your output first! Expect to check and edit content before it goes live!

Slide 50

Slide 50 text

pa.ag @peakaceag 50 "As an AI language model" query reveals just some of the many issues (e.g., lack of quality control, fakes, etc.) Sources: https://pa.ag/3nuowxU & https://pa.ag/3ALso0B

Slide 51

Slide 51 text

AI for: Audio

Slide 52

Slide 52 text

pa.ag @peakaceag 52 beatoven.ai: Create customisable royalty free music Beatoven.ai uses AI music generation to compose unique mood-based music to suit every part of your creative piece, e.g., video (can be drag and dropped while creating): Source: https://pa.ag/3T9xQD4

Slide 53

Slide 53 text

pa.ag @peakaceag 53 More AI audio & music tools Krisp AI: Removes background voices, noises and echo from all your calls Cleanvoice AI: Removes filler sounds, stuttering and mouth sounds from podcasts or audio recordings Podcastle AI: Studio-quality recording, editing, and seamless exporting – in a single web platform soundraw.io: Royalty-free music, AI generated based on mood, genre and length. Descript.com: To write, record, transcribe and edit video and podcasts Soundbite AI: Turn new and existing audio and video content into ready-to- edit social media posts, and summaries.

Slide 54

Slide 54 text

AI for: Video

Slide 55

Slide 55 text

pa.ag @peakaceag 55 Want to boost personalised interactions with customers? Maverick records once but automatically personalises for all, enabling videos at scale. Source: https://pa.ag/3ZSIZdu

Slide 56

Slide 56 text

pa.ag @peakaceag 56 The Peak Ace team used MURF.ai for a cost-effective localisation of video ads assets: Step 1: Transcribe original script and localise into desired language Step 3: Input your localised copy and export script Step 2: Decide on the voice and style you'd like to use Step 4: Add the new script to your video and adjust until it sounds perfect New FR version Our task/challenge: A client wanted to use a video in different regions but didn’t have local employees to voice the scripts. Original DE

Slide 57

Slide 57 text

pa.ag @peakaceag 57 Other video AI solutions worth checking out:

Slide 58

Slide 58 text

pa.ag @peakaceag 58 Gen-2 by Runway in particular looks crazy cool: Just look at how it transfers the style of any image or prompt to every frame of a video: Source: https://pa.ag/41WwvTp Source Video Generated Video Driving Image

Slide 59

Slide 59 text

AI for: (More) Productivity

Slide 60

Slide 60 text

pa.ag @peakaceag 60 Compose AI: A Chrome extension to make you fast(er) A free Chrome extension to automate your writing Source: https://pa.ag/41TToay

Slide 61

Slide 61 text

pa.ag @peakaceag 61 MailMaestro: Bringing OpenAI LLMs directly to Outlook Generate custom emails and responses based on short prompts; create one click email thread summaries including action points for the receiver: Source: https://www.maestrolabs.com

Slide 62

Slide 62 text

pa.ag @peakaceag 62 Create more, better LinkedIn content with AI using Taplio Establishing a personal brand on LinkedIn is a key pillar of growth. Stop wasting endless hours writing your next post and get fresh ideas using Taplio’s content inspiration layer: Source: https://taplio.com

Slide 63

Slide 63 text

pa.ag @peakaceag 63 ExcelFormulaBot: turn your problem into a formula Transform your text instructions into Excel formulas in seconds (yep, I know: ChatGPT can do that as well…) Source: https://pa.ag/3ythYlb

Slide 64

Slide 64 text

pa.ag @peakaceag 64 Otter.ai for automated meeting notes & summaries Your meeting assistant that records audio, writes notes, automatically captures slides, and generates summaries: Source: https://pa.ag/3LhCcWC Word of advice (and caution, again): It's super practical but getting consent to record is important – especially since some services can automatically send meeting summaries by email afterwards! Also, I am not a lawyer.

Slide 65

Slide 65 text

AI for: No Code

Slide 66

Slide 66 text

pa.ag @peakaceag 66 Akkio.com: Grow faster with No-Code ML Create custom ML models in minutes without a technical background: Source: https://pa.ag/3JwknSC

Slide 67

Slide 67 text

pa.ag @peakaceag 67 Things Peak Ace has done with Akkio and their models: 1. Sales forecasting – e.g., understanding when customers are likely to buy. 2. Churn prediction – e.g., finding out which customers are likely to cancel next. 3. Lead scoring – e.g., understanding which sources generate the highest purchase intent.

Slide 68

Slide 68 text

AI and: ChatGPT

Slide 69

Slide 69 text

pa.ag @peakaceag 69 ChatGPT is currently using OpenAI’s GPT-4 model Including capabilities for browsing (Bing) as well as plug-in usage Source: https://chat.openai.com/

Slide 70

Slide 70 text

pa.ag @peakaceag 70 What makes a good prompt and how do you structure it? Some tips to create better and more efficient prompts for ChatGPT (h/t Mike King): Source: https://pa.ag/42KbNpH Role: Who is ChatGPT creating as? Context: What is the situation it is creating for? Instructions: What specifically do you want it to do? Format: How do you want it to return its response? Examples: Samples of the output that you expect for pre-tuning. Constraints: What should ChatGPT not do?

Slide 71

Slide 71 text

pa.ag @peakaceag 71 ChatGPT combined with AIPRM (prompt mgmt. plug-in) Access a curated selection of ChatGPT prompts specifically designed for SEOs, marketers, sales, support, copywriting and more : Source: https://aiprm.com/

Slide 72

Slide 72 text

pa.ag @peakaceag 72 ChatGPT browsing mode using Bing search: Essentially, OpenAI uses Bing search to enrich output based on automatically generated queries taken from the original ChatGPT prompt:

Slide 73

Slide 73 text

pa.ag @peakaceag 73 ChatGPT plug-ins for more productivity/efficiency e.g., LinkReader allows direct interaction with PDFs, Word documents, etc.

Slide 74

Slide 74 text

Direct integrations into LLMs through plug-ins are going to significantly change the web’s ecosystem

Slide 75

Slide 75 text

pa.ag @peakaceag 75 The new kid (concept) in town: Auto-GPT “The main feature is the ability to use ChatGPT to independently create prompts to plan how to complete a task and then create more prompts for finishing that task.” Source: https://pa.ag/3oWAuRr If the AI agent finds itself unable to complete the task, it will create new prompts to figure out how to proceed. [… ] Auto-GPT is self-prompting, removing the need for creative and detailed prompts. All it needs is a set of goals for a task to complete.

Slide 76

Slide 76 text

pa.ag @peakaceag 76 Automation at scale through OpenAI’s APIs Give it a try using the OpenAI Playground for building/testing your own apps: Source: https://platform.openai.com/playground/

Slide 77

Slide 77 text

pa.ag @peakaceag 77 Google Ads script: RSA generation based on API calls Leverage OpenAI’s API to use the maximum number of responsive search ads assets and, in turn, boost your paid search campaigns: Source: https://pa.ag/40GipUM & https://pa.ag/3phXv1B Another interesting idea:

Slide 78

Slide 78 text

AI vs. Search

Slide 79

Slide 79 text

pa.ag @peakaceag 79 The "new Bing" – Microsoft going all-in on AI in search Bings’ integration of OpenAI’s LLM into its search engine kicked-off a whole series of changes for search in general (some changes are still to come):

Slide 80

Slide 80 text

pa.ag @peakaceag 80 It’s still early days, but initial ad integrations are available How will targeting work? What’s the model? How about bidding? So many questions … Source: https://pa.ag/3Ngtk4C

Slide 81

Slide 81 text

pa.ag @peakaceag 81 Naturally, Google has something brewing as well… Meet Google Bard (which is not officially available in Europe just yet…) Source: https://bard.google.com/ Just use a VPN service of your liking and set your IP origin to the US, or Switzerland even.

Slide 82

Slide 82 text

Search Generative Experience; supercharging search with generative AI Google SGE

Slide 83

Slide 83 text

pa.ag @peakaceag 83 Google‘s Search Generative Experience (SGE) Join the waitlist (if you’re not already on it): g.co/Labs (US proxy needed) Source: https://pa.ag/433w36x

Slide 84

Slide 84 text

pa.ag @peakaceag 84

Slide 85

Slide 85 text

pa.ag @peakaceag 85

Slide 86

Slide 86 text

pa.ag @peakaceag 86

Slide 87

Slide 87 text

Your marketing team needs to be thinking about adapting their SEO strategy, as obtaining organic traffic is likely to get much more difficult in the future. Why should you care?

Slide 88

Slide 88 text

Organic search results will change, but so will ad creation and the integration of ads into the search results. Google to introduce generative AI for ads as well

Slide 89

Slide 89 text

So, where are AI & LLMs headed in the future?

Slide 90

Slide 90 text

pa.ag @peakaceag 90 Amazon entered the game with impressive initial results Amazon has released a new language model that outperforms GPT-3.5 on ScienceQA by 16 percentage points, or with 75.1% to 91.6% accuracy Source: https://pa.ag/3Fl5ZKe Amazon researchers came up with Multimodal-CoT, which combines visual features in a separate training framework, to reduce the effects of these mistakes [hallucinatory reasoning patterns]

Slide 91

Slide 91 text

pa.ag @peakaceag 91 And their own on-site search is most probably next: Source: https://pa.ag/3WUrTfb

Slide 92

Slide 92 text

pa.ag @peakaceag 92 Language models using APIs for even better results This paper shows that LMs can teach themselves to use external tools via simple APIs Source: https://pa.ag/3T1Nq3z Toolformer: A model trained to decide which APIs to call, when to call them, what arguments to pass, and how to best incorporate the results into future token prediction.

Slide 93

Slide 93 text

With one significant goal in mind: improving themselves Models will generate their own training data

Slide 94

Slide 94 text

Because today’s LLMs make stuff up Models will fact-check themselves

Slide 95

Slide 95 text

What if a model could use only the most relevant subset of its parameters to answer a query vs. running through every single one of its parameters instead? Massive sparse expert models?

Slide 96

Slide 96 text

pa.ag @peakaceag 96 MS introduced Kosmos-1, their first multimodal model Kosmos-1 can reportedly analyse images for content, solve visual puzzles, perform visual text recognition, pass visual IQ tests & understand natural language instructions: Sources: https://pa.ag/3L9JP1d

Slide 97

Slide 97 text

True multimodality is going to be a significant step in AI evolution!

Slide 98

Slide 98 text

pa.ag @peakaceag 98 Planning for Artificial General Intelligence and beyond “As our systems get closer to AGI, we are becoming increasingly cautious with the creation and deployment of our models.” Source: https://pa.ag/3mxjSOU The first AGI will be just a point along the continuum of intelligence. We think it’s likely that progress will continue from there […] If this is true, the world could become extremely different from how it is today, and the risks could be extraordinary. A misaligned super intelligent AGI could cause grievous harm to the world; an autocratic regime with a decisive superintelligence lead could do that too.

Slide 99

Slide 99 text

pa.ag @peakaceag 99 The best part of this OpenAI post though? Source: https://pa.ag/3mxjSOU Successfully transitioning to a world with superintelligence is perhaps the most important— and hopeful, and scary—project in human history. Success is far from guaranteed, and the stakes (boundless downside and boundless upside) will hopefully unite all of us.

Slide 100

Slide 100 text

...hopefully.

Slide 101

Slide 101 text

pa.ag @peakaceag 101 Meanwhile: World’s first AI-flown fighter jet can dogfight In a joint project between DARPA and the US Air Force, a special aircraft called the X-62 Vista became the first tactical aircraft to be piloted by AI: Source: https://pa.ag/3JtRtlS The jet was under the control of one of four different AI algorithms at any given time during the tests. […] This included dogfighting in simulated combat missions, as well as takeoffs and landings

Slide 102

Slide 102 text

Are we really ready for this?

Slide 103

Slide 103 text

pa.ag @peakaceag 103 How do we put this together and how do we think about implementing AI to grow our business? Some ideas: 1. Automation = efficiency, the more menial tasks you automate, the more time you and your team have to tackle the bigger picture & creative items. Don’t forget individual tasks such as emails or meetings, either. 2. Budgets are tight. Leveraging AI tools for content, ad or image creation and other tasks to tackle areas that usually require additional resources result in cost savings across multiple areas of a business. 3. Utilise tools like Akkio to understand the areas of opportunity within your existing business. Identify top prospects to target and go after to drive high conversion rates.

Slide 104

Slide 104 text

Bastian Grimm www.pa.ag twitter.com/peakaceag facebook.com/peakaceag bg@pa.ag Peak Ace: Delivering Digital Experiences Questions? Thoughts?