Random Observation/Comment #820: My digital twin is more than just identity – it’ll be the sculpted version of “me” with the ability to create on my behalf.
Why this List?
AI is still accelerating. At top of mind, we’re seeing large existing companies roll out their optimizations directly to their existing SaaS customers. This distribution of tools leveraging AI is going mainstream very quickly. We’ll likely see a proliferation of a new narrative of “organic” non-AI generated content, but it’ll be temporary to the mass flooding of images, videos, and posts shared by bots.
The AI relevant posts:
- 30 ChatGPT Musings and Use Cases (Dec 2022)
- 30 AI and AGI Trends (Apr 2023)
- 30 Practical AI Tools (May 2023)
- 30 Generative AI Image Experiments (Jun 2023)
- 30 Common AI Requests (Jul 2023)
- 30 ChatGPT Future Features (Aug 2023)
- 30 Ideas for an Applied AI Working Group (Sept 2023)
- Text-to-Anything – The “What You Narrate is What You Get” (WYNIWYG) is not going away. There’s so much crazy advancement with batching of inputs and outputs for full automation.
- “Magic” Studio for everything media – Canva’s latest release is clear that image generation/alteration will directly be shared with creators and advertisers
- All Internet content will become AI-generated/altered – The boom of content will be as significant as writing letters vs writing emails
- Conversations instead of Search – Search Generative Engines (SGE) leads to finer tuning of requests in multiple stages of conversation rather than finding the perfect single search
- ChatGPT Enterprise has launched – I think the private training of LLMs on business products will lead to new valuations of companies based on their trained AI
- ChatGPT Enterprise for Mergers and Acquisitions (M&A) – If I’m acquiring a new company, the “historian” type of roles may be less important if you could communicate your product’s LLM to the old company’s LLM. This could be the start of LLM collaboration, migration, and interoperability. AI Interoperability might be super simple because the AIs would just talk to each other and ask each other questions recursively.
- AI Agents as natural language APIs – Bard Extensions connecting to Workspace and initiating searches for Google Arts and Culture, Google flights, or Google maps connects everything together.
- Expanding voice-activated services like Google Assistant, Siri, and Amazon Alexa to more conversational interactions – This is not just interpreting voice-to-text-to-OS commands/Search results. This would provide personalized knowledge of existing services. I think this gets super powerful if you can review drafts or activities prior to post.
- Voice-first operating systems – It might be easier to remove screen time by making more things audio-specific
- Multi-language Everything – YouTube Creator AI Tools roadmap and Spotify AI Voice translations of audio-to-audio podcasts is going to be available. Your content will find dozens of new audiences and cultures. Heygen can also do direct video translations and match your mouth movements.
- AI Agents joining chats – MetaAI has launched these on Facebook Messenger and Whatsapp already.
- AI Agents working together (ChatDev) – Two Minute Papers does an incredible breakdown of this and at the end of the video sneaks in some brain explosions about AI Agents creating games that can host AI Agents.
- Video Game Simulations recursive “Inception” – If you ask your AI Agent company to write its own video game that they can deploy AI Agents, then you can recursively create more focused video games on optimizing some application. Imagine we’re in the simulation optimized for finding meaning in life and we needed to create AI bots.
- Wearables entering the trend – If your devices are already listening all the time then maybe it can add context and start to connect your conversations to notes or actions.
- Migration to the Metaverse – Text-to-image-to-3D for second Earth creation off of crowd-sourced photos. The NVIDIA Flexicube is crazy cool.
- Prompting the LLMs to help us prompt – This is a great prompt engineering trick in general. If you don’t know what context you need to provide, just ask! Set the goals and GPT4 really does an incredible job of completing these tasks. I keep on pushing to ask more and get deeper into the details.
- GPT-4V(Vision) – Bard already has this feature of uploading images to get descriptions, but it doesn’t look like it was testing for OCR + additional logic and context. The GPT-4V testing I’ve seen on examples have been incredibly complex. I’m just looking at Twitter/X on the #GPT-4V and you can see some pretty incredible use cases
- Ethical and psychological impacts – Beyond misinformation and deep fakes, I am a little mentally distraught on what the point of my digital life becomes if it can all be faked. There was a recent video of a guy, Kyle Vorbach, faking his life on social media for a month.
- Regulatory Concern and impact on Intellectual Property – Creators in Hollywood are doing 3D scans of themselves so they can sue people for using their face and images.
- Future of Cloud in AI – Cloud access to LLMs is going to be such a huge infrastructure cost to hosting these services.
- Closed Source vs Open Source / Retail vs Dark Web – The largest LLM is closed source private, GPT4, with 1.7T parameters. The conversation here is that Closed Source LLMs will include guardrails to prevent the negative prompting. It’s like search engines preventing the Dark Web of pages from getting through to the regular retail user. I think the “Dark Web” of LLMs will be open source and likely live in some seedy use cases.
- Training LLMs with other LLMs – Other LLMs can train via querying large LLMs in order to create smaller parameters with similar performance. Tiny LlaMa, for example, is only 1.5B parameters. The interesting techniques of training LLMs is to ask the “Teacher Agent” for the answer and also the specific reasoning behind the answer.
- Smaller LLMs that can fit into a smart phone – If the LLM doesn’t need to know the whole internet, perhaps we can have a smart enough LLM that can fit in memory of a smart phone. This would then be your personalized agent that knows your schedule and daily tasks.
- Private data connectivity and permissions to AI Agents – If AI agents go further with connectivity, I think we would want to segregate my knowledge of an internal day-to-day Clembot with access to my Gmail and Calendar versus an external facing Clembot that can be queried for work assignments.
- The “AI Stigma” – Is it okay to use AI on your everyday tasks? There’s a breadth of use cases, but would someone see the AI-generated content to be not as good or pushing towards the AI overlords replacing our jobs? This stigma is real and will divide people into the embracers/evangelists and the “organic”/non-GMO/non-AI creators.
- Writing longer form and then Synthesizing into briefs – It might happen sooner than later, but it would be crazy if a sender writes in bullet points so that their AI can convert it to business email, and then the receiver’s AI reads and converts it back to bullet points.
- Training on your search results – Google is indexing Bard searches, so nothing is private.
- Change of perception on Internet Truth – We already think most information is promoted advertisement or some type of misinformation. Now that all these tools are retail-available for no cost, we’ll just see a larger explosion of fake content. We’ve already surpassed all 150B+ photos taken by humanity with AI generated content.
- AI Gaming interactions – AI NPCs are growing quickly and I think it’ll be really interesting to see how it changes the types of games we play. I would also love a live MetaAI assistant within my Quest 3.
- GPT as your digital successor – This was also my 30th number with “Clembot from the grave”. I really think applying a Clembot LLM to become a successor trained to provide decisions for a Trust would be super interesting. Clembot Chief Financial Officer and Risk manager will grow my Trust portfolio consistently.
~See Lemons Follow AI