Welcome back to SundAI ! – 🎆 last edition of 2024 🎇

This week in AI was a black hole of chaos, with a lotta tech confetti and lit up with a flamethrower full of ambition, because OpenAI hit us with their latest “o3” model (which you cannot play with yet, if you aren’t one of their guinnea pigs), Microsoft cracked open its PromptWizard (sounds fun!), Instagram decided “deepfakes for everyone” is the next best thing, and Google expanded Gemini’s linguistic tentacles. And Apple, being late at the AI party, is now on track to become the first $4 trillion company (sick!), because why not, and Elon Musk continues to stir the AI pot like it’s his personal cauldron.

So burn your overpriced subscription plans, take a very, very stiff drink, and let’s go down into what happened in the void of the last week of 2024!

If you have the attention span of a goldfish, here’s your TL;DR.

OpenAI’s o3 model blew past human-level performance in coding and math, but its price tag makes it more of a birthday gift to someone else, than a solution for everyday problems. And Thelonius Musk’s xAI raised $6 billion to hoard GPUs for its Colossus supercomputer. Would this be for world domination or just trolling at scale, like he usually does, but now with more GPU power. And Apple just hit $4 trillion in value. That is a four with twelve zero’s, or 4.000.000.000.000 USD. To give you an idea, Germany has a Gross Domestic Product of 4.5 trillion. And for that amount of money you get all the Mercedes, and BMW’s to drive in, plus the entire population to help you get around. And that is all because of it’s latest innovation – Genmoji ! But let’s face it, Siri’s glow-up still can’t order pizza without a meltdown.

Instagram’s came up with “Movie Gen” tools that let you AI-edit videos into perfection, which is making your friend’s Pattaya vacation (Thailand, yes) even more suspicious. Google’s Gemini 2.0 Flash Thinking Mode now can speak 40 languages, because why not let AI out-debate you in Mandarin too? Jadve and YouTube Dream Screen launched tools for instant AI video making, and custom designed backgrounds. Synthesia’s AI avatars are taking over video presentations. It is a bit better than HeyGen, but nonetheless worthless compared to the real deal.

GitHub did a good thing for society – well at least for AI coding – because they brought Copilot Free to help to the masses, but it has enough limits to remind you that it’s still worse than your actual hooman co-worker. Anthropic teased its Claude-powered Cookbook. Though it sounds tasty, it’s basically a toolkit that is designed to help developers build and refine AI applications using Claude. It has a few pre-built code snippets and templates for implementing AI in various tasks. Yippee. Not.

Salesforce gave us Agentforce 2.0, which is designed to let bots invade your workspace, particularly Slack, and steal your thunder with it’s pre-built skills and workflow integrations. Why would you ask? Dunno, just look it up yourselves. It’s got something to do with automating the daily mundane stuff. Yeah, like your work. So basically not revolutionary. What is a good development though, is that Google hinted at AI deepfake detectors, which are going to make the internet just a tad safer while everyone else still embraces the chaos of fakery. And last, and certainly absolutely least, Apple gave us….. Image Playground. Yes, yet another generative AI art tool that no one asked for, which is killing the planet. So don’t expect anything Picasso-level unless you really hate Picasso.

And now, ze commercials

---

If you like this article and you want to support me:

1. Comment, or share the article; that will really help spread the word 🙌
2. Connect with me on Linkedin 🙏
3. Subscribe to TechTonic Shifts to get your daily dose of tech 📰

---

SundAI // Week 52 – AI’s end-of-year chaos parade

Neural updates:

1. OpenAI o3: The “just enough to panic” model
2. Elon Musk’s $6B fallus enlarger
3. Apple’s $4 trillion flex
4. Instagram goes full deepnude
5. Google expands Gemini’s language wizardry
6. Microsoft PromptWizard. Apparently Microsoft thinks prompts still matter
7. Salesforce is here with Agentforce 2.0

---

AI tools you didn’t know you needed

Here’s the latest batch of AI tools to ruin your productivity for y’all wannabe creators and procrastinators out there:

• Synthesia: Create lifelike AI videos with avatars. They could be so lifelike that it could replace you at your next work presentation. But the audio/video combo could improve. Suggested use case: Panic about your job while it codes circles around you.
• Canva Magic Media: Turn text into video, no talent required. Perfect for those “I woke up like this” Instagram reels. Suggested use case: Panic about your job while it codes circles around you.
• Jadve: Automates video creation, from script to final cut. Warning: may make actual video editors obsolete. Suggested use case: Panic about your job while it codes circles around you.
• YouTube Dream Screen: Add AI-generated backgrounds to your clips, because why bother with real scenery when pixels will do? Suggested use case: Panic about your job while it codes circles around you.

---

Tools

1. Anthropic Cookbook: A one-stop recipe book for all things Claude, offering code and guides to help developers cook up their own AI-powered applications. Perfect for those who think Claude’s not spicy enough straight out of the box.
2. Genesis: The Swiss Army knife of robotics platforms, designed for general-purpose embodied AI and physical AI projects. Think of it as a gym membership for your robot that does more than just flex.
3. Picotron: A no-frills repo for pre-training Llama-like models with 4D Parallelism. If you want lean, mean LLMs without the bloat, this is your dojo.
4. Helicone: An open-source observability platform for LLMs. Because you can’t fix what you can’t see, and this tool is like giving your AI a pair of glasses and a reality check.

---

Something to pretend you’re working

1. Alignment Faking in Large Language Models: Turns out, your chatbot might be gaslighting you. This paper explores how LLMs fake “agreeing” with you while secretly rolling their virtual eyes.
2. AI Safety on a Budget: A survival guide to implementing safer LLMs without needing a billionaire’s wallet. Free tools, guardrails, and datasets to keep your bot from going rogue.
3. Fine-Tuning LLMs for RAG: A must-watch for today’s AI tinkerers. Learn how to make your LLMs play nice with retrieval-augmented generation systems without breaking a sweat.
4. The Real Reason Your Company’s AI Isn’t Working: Spoiler: It’s not the tech; it’s your company’s dinosaur-like resistance to change. This article dives into the organizational hurdles killing AI adoption faster than you can say “budget cuts.”
5. Introducing ReACT LLM Agents: Meet the AI agents that can think and act at the same time. This concept might just make your assistant smarter than your manager.

---

Top research papers of the Week

1. Qwen2.5 Technical Report: Qwen just got buffed up with a pre-training dataset scaled to 18 trillion tokens. It’s like taking an already ripped model and slapping it into overdrive with supervised fine-tuning. Meta, you’ve been warned.
2. Byte Latent Transformer: Patches are the new tokens! This byte-level architecture improves inference efficiency and robustness, breaking new ground for handling raw, unstructured data without choking.
3. Deliberative Alignment: OpenAI’s o-series models just got a moral compass. This new training method makes LLMs safer by teaching them to reason explicitly through a chain of thought.
4. Fully Open Source Moxin-7B Technical Report: Moxin’s zero-shot capabilities outperform many larger models while sticking to an open-source ethos. Finally, a 7B model that walks the open talk.
5. RAGBench: Benchmarking retrieval-augmented generation systems with 100k examples sourced from real industry corpora. This is where theory meets hands-on practicality.
6. CosyVoice 2: Streaming speech synthesis just leveled up. CosyVoice 2 combines LLM-powered improvements with better utilization of speech tokens for more natural, scalable voice applications.

---

Links for lazy folks

• 1-800-CHATGPT: Chat with AI over the phone, just like it’s 1998 again.
• GitHub Copilot Free: Code smarter, not harder, now with limits because life isn’t fair.
• Anthropic Cookbook: Whip up some Claude-powered AI magic, no aprons needed.

So, that’s finally it for this week, and this year.

I am wrapping up 2024 with a sigh and a laugh, and enough existential dread to keep me awake at night. End of 2024, I can conclude that AI is not slowing down, and neither am I. 2025 still strong with one post a day, if God is willing. So stay tuned, stay sarcastic, and let’s see what fresh chaos 2025 has in store for the only Big Tech Tabloid.

Signing off from the frontlines of AI insanity, where Siri’s still useless, and Grok can’t decide if it’s a bot or a comedian.

Marco

---

Well, that’s a wrap for today. Tomorrow, I’ll have a fresh episode of TechTonic Shifts for you. If you enjoy my writing and want to support my work, feel free to buy me a coffee ♨️

---

Think a friend would enjoy this too? Share the newsletter and let them join the conversation. LinkedIn appreciates your likes by making my articles available to more readers.

To keep you doomscrolling 👇

1. Brace, brace brace! AI takes the stick at Heathrow’s air traffic control center | LinkedIn
2. AI is a compulsive liar | LinkedIn
3. In 2025, AI needs to put up or just shut up! | LinkedIn
4. A 17 yo brat created a $1M/month app. Here’s how he did it. | LinkedIn
5. This is a eulogy for chegg. Gone but not forgotten (unless you’re a student, then definitely otten) | LinkedIn
6. Musk wants to make games great again | LinkedIn
7. The great tech wake-up call: Developers, meet the dystopia you helped build | LinkedIn
8. Flamethrower dogs, kamikaze cars, and bomb-planting humanoids. | LinkedIn
9. Objection! Your honor, ChatGPT made me do it | LinkedIn
10. A cautionary tale about an AI unicorn that turns into a fraudulent little pwny | LinkedIn
11. Meet Daisy, the AI Granny who’s here to waste scammers’ lives | LinkedIn
12. AI Search Engine Optimization | LinkedIn
13. I’ve seen the dark side of AI, and you need to know about it | LinkedIn