Howdy! It’s Barsee again.

2024 was a big year for our newsletter!

We crossed the 100,000 subscribers milestone, wrote 139 emails, and exceeded 16M+ views! It gave us the opportunity to connect with an incredible audience. This year, we covered so many exciting AI releases, groundbreaking updates, and innovative products, and we couldn’t have done it without your support. Thank you for reading, engaging, and growing with us!

PREDICTIONS

12 Predictions for AI in 2025

The past year in tech has been one of steady and meaningful progress.

AI technology improved a lot, with better memory, longer attention spans, faster processing, easy access to video generation, and smarter systems overall. Other fields like self-driving cars, virtual reality, brain-computer connections, and quantum computing also took big steps forward, bringing fresh excitement to the tech world.

And as we are heading into 2025, here’s some of the predictions that I think we will see:

• Google will take the lead on real-life image and video intelligence due to its vast datasets from Google Images and YouTube.

• AI will start to be seriously used in AI development, starting a smaller form of recursive self-improvement, though I don't expect it to be without human intervention in 2025.

• Web agents will go mainstream, becoming the next major killer application in consumer AI. Imagine a world in which your web interactions will be done by an AI web agent. Claude’s computer use and Open AI have something lined up in January.

• It will become common place for all AI models to have token limits higher than 10M with the best models having near-infinite memory (i.e. in the hundreds of millions of tokens with 99.9% recall accuracy.)

• The real story of 2025 won't be companies giving up on GenAI - it'll be companies finally figuring out how to use it right. (Having experimented with AI since 2022, it’s all about finding the right use case.)

• AI video indistinguishable from real videos akin to current AI images + A semi-decent full-length AI-generated movie.

• A somewhat playable AI video game i.e., a superior version of the diffusion-based video game we saw in the deepmind paper a few weeks back.

• Self driving cars will be better than humans.

• Real general intelligence won't be archived in 2025 either. It will take some time for General Intelligence in AI. AGI got captured as a marketing term by OpenAI in 2022.

• Massive anti-AI sentiment spreading like wildfire.

• Deepfake for both images and videos will improve massively to the point it gets impossible or really hard to distinguish.

• We see some wild combinations - imagine brain-computer interfaces working alongside everyday AI tools. That's where the real magic will happen.

PS: take them with a grain of salt, folks – the future of AI is always full of surprises!

TRENDING TOOLS

• Betterwatchlist > The stock market watchlist with AI superpowers. Get AI-powered insights on why prices are moving in real time. (link) *

• Chance AI > Snap a photo to unlock a world of information, context, and hidden narratives. (link)

• Gitdiagram > Turn any GitHub repository into an interactive diagram for visualization. (link)

• Prompt Improver > Conversational AI agent for prompt engineering. (link)

Wait, I have more for you today! Here's an image featuring some of the best AI tools of 2024 for every type of work. (image credit goes to A16z)

2024

Now let's rewind the top AI projects of the year!

OPENAI

• o3 & o3-mini (announced): Advanced reasoning models for coding, math, and science. o3-mini allows users to adjust reasoning time.

• o1: A new series for problem-solving tasks like code generation and document comparison.

• GPT-4o: Received updates for structured outputs and better speech/audio interaction.

• ChatGPT Search: It provides a faster, more accurate web search experience with direct links to high-quality sources, perfect for up-to-date news, sports scores, stock quotes, and more.

• Sora: A text-to-video model for generating video content from text.

• Advanced Voice Mode: Enhances voice assistant interaction.

GOOGLE

• Gemini 2.0: It can generate human-like speech, create images with text, invoke functions, and enable real-time interactions with text, audio, and video, while also understanding spatial and video content for summaries or overviews.

• Veo 2: High-definition video generator from text. It promises to deliver more realistic videos with a better understanding of physics and human movement.

• Imagen: Text-to-image model with better realism and resolution.

• Project Astra: A multimodal AI assistant that can interpret visual and audio inputs in real-time, identify objects, locate misplaced items, and explain code.

• NotebookLM: Document-synthesizing tool with audio summaries.

• LearnLM: AI tutors for personalized education.

• SynthID: Watermark tool to identify AI-generated images.

META

• Llama 3.2: Meta's Llama 3.2 incorporates new medium-sized vision LLMs (11B and 90B), along with lightweight, text-only models (1B and 3B), that can run on mobile devices.

• Orion AR Glasses: Holographic AR glasses with voice, eye, and hand tracking.

• Meta AI Assistant: AI assistant built with Llama 3, offering users a conversational AI experience across various platforms.

• Rayban Meta Glasses: Smart glasses with multimodal AI for real-time translation and accessibility.

TESLA

• Cybercab: A new electric vehicle dedicated to self-driving that lacks a steering wheel or pedals.

• The Robovan: It features a sleek design and self-driving capabilities, capable of transporting up to 20 passengers or a large cargo.

• Optimus personal assistant robots: Tesla claims that these robots can mow lawns, fetch groceries, and babysit, estimating their cost to be between $28k and $30k.

APPLE

• Apple Intelligence: A personal intelligence system that uses generative models to help users communicate, work, and express themselves on their iPhone, iPad, and Mac.

• Enhanced Siri: More natural voice and better conversational abilities using ChatGPT Integration.

ALIBABA

• Qwen 2.5: Open-source models with versions up to 72 billion parameters, excelling in math, coding, and multilingual tasks.

ANTHROPIC

• Claude 3 Model Family: Released in March, this family includes Haiku, Sonnet, and Opus models. Each is designed for different needs, allowing users to choose between intelligence, speed, and cost.

• Claude 3.5 Models and Computer Use Feature: Claude 3.5 models offer enhanced capabilities, including a feature for interacting with a computer’s graphical interface to perform tasks like web searches and typing with user permission.

Let me know if I missed anything.

ICEBERG

The AI Iceberg of 2024

The diffusion of AI through society and the economy is like the proverbial iceberg - what we observe is merely the visible tip, while the majority remains hidden beneath the surface.

Some speculative observations:

Unofficial AI Usage: The majority of AI utilization “at work” happens unofficially and goes unrecorded. Workers quietly and nonchalantly incorporate tools like ChatGPT into their daily workflows because it helps them do their job better and the cost can be rounded down to zero. For every worker using their company’s sanctioned and deeply integrated “copilot” there are thousands who JFDI with their own provisions.

Opaque AI Lab Developments: Leading AI laboratories maintain secrecy around their research and development. We only know what they release, and even though the pressure to release fast is significant, there are hints that much more goes on behind closed doors. This is true even for labs releasing “open” models.

Invisible AI Content: While poorly generated AI content is easily identifiable, high-quality AI-produced materials like text and images often goes undetected. We only notice AI's failures, not its successes. A lot more content you encounter every day is AI-generated than you might imagine.

Silent Business Integration: Many companies integrate AI into their systems and processes without public disclosure. They do this to preserve their competitive advantage, and to avoid public criticism, sometimes justified and in other cases not, of their reliance on AI.

HUGGINGFACE

Two of the most loved Huggingface space of 2024

Kolors Virtual Try on: It allows you to try on clothes before purchasing, ensuring a perfect fit every time.

Illusion Diffusion: It allows you to generate stunning high quality illusion artwork.

THINK PIECES / RESOURCES

• ChatGPT search vs. Google: A deep dive analysis of 62 queries. (link)

• AI traps: Lessons from launching Webflow's AI Assistant. (link)

• 60 of Google’s biggest AI announcements in 2024. (link)

• How to use Al to become a learning machine. (link)

CONTENT CORNER

THAT’S ALL FOR TODAY

That’s all for today’s issue, folks. And Happy New Year!

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ Like what you see? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.