LLMs - 2025 Year in Review

Hi everyone,

Andrej Karpathy (one of the founders of AI) made a sort-of “year-in-review” of LLMs in 2025. It’s worth going through that.

Also, we’re planning something big for the newsletter in 2026. Stay tuned.

Let’s get right into it.

In this issue:

• 🤝 In Partnership: AI sidehustles that make money

• 🤿Deep Dive: Karpathy’s 2025 LLM Reality Check

• 🤝 Powered by: Trade alerts before market moves

• 💬 TIW Office Hours: Get your questions answered

• ⚒️Tool Snapshots: Tools for AI, no-code, and productivity

• 🖼️AI Art: Examples of great and trending AI art

• 🤝 Supported by: Scale support with AI

🤝IN PARTNERSHIP WITH THE HUSTLE DAILY

200+ AI Side Hustles to Start Right Now

While you were debating if AI would take your job, other people started using it to print money. Seriously.

That's not hyperbole. People are literally using ChatGPT to write Etsy descriptions that convert 3x better. Claude to build entire SaaS products without coding. Midjourney to create designs clients pay thousands for.

The Hustle found 200+ ways regular humans are turning AI into income. Subscribe to The Hustle for the full guide and unlock daily business intel that's actually interesting.

Start making AI money

🤿 DEEP DIVE

Karpathy’s 2025 LLM Reality Check

Intelligence: 2025 reshaped how large language models are trained, used, and understood, shifting progress from bigger models to smarter optimization, new app layers, and radically different interaction patterns. Karpathy (one of the most respected voices in AI) summarized his findings.

• RLVR replaced “just pretrain more” as the main capability lever. Reinforcement Learning from Verifiable Rewards let labs push reasoning via long optimization runs and test-time compute, diverting massive resources away from raw pretraining and into post-training gains.

• LLM intelligence revealed itself as jagged, not general. Models spike in domains with verifiable rewards while remaining fragile elsewhere, undermining benchmark trust and exposing how easily systems overfit to test-like environments.

• A new application layer emerged, led by tools like Cursor. These apps orchestrate multiple model calls, manage context, and embed humans in the loop, suggesting labs train “general grads” while apps turn them into working professionals.

• Agents moved onto personal machines with Claude Code. Running agents locally, close to files, tools, and low-latency feedback, proved more practical than cloud-first autonomy for today’s uneven capabilities.

• “Vibe coding” crossed the threshold into real productivity. Natural language became a viable way to create disposable, exploratory, and highly customized software, changing who can build and why code gets written at all.

• Interfaces started shifting beyond chat with Gemini Nano Banana. Multimodal outputs like visuals, layouts, and apps hint at an LLM GUI era where text is no longer the primary human interface.

Bottom line: progress in 2025 was less about smarter models in theory and more about where intelligence is applied, how it is packaged, and who can actually use it. The gap between capability and real-world leverage remains wide, which is exactly where opportunity sits.

🤝POWERED BY ELITE TRADE

If You Could Be Earlier Than 85% of the Market?

Most read the move after it runs. The top 250K start before the bell.

Elite Trade Club turns noise into a five-minute plan—what’s moving, why it matters, and the stocks to watch now. Miss it and you chase.

Catch it and you decide.

Get the Next Alert

_{By joining, you’ll receive Elite Trade Club emails and select partner insights. See Privacy Policy.}

💬 TIW OFFICE HOURS

Q&A

Q: What happens to junior roles by 2026?

A: The floor rises. New hires get productive faster because AI fills gaps. But expectations also rise. “Learning by doing” becomes “learning by reviewing and correcting.”

Q: Who will win in 2026? OpenAI, Gemini, Anthropic?

A: No idea… OpenAI won consumer market and it will likely stay that way. Gemini has distribution through emails, calendars, etc. Anthropic is the king with coding and enterprise. So, it’s tough to answer, they’ll all win their different battles. Admittedly though, I just don’t see how OpenAI is worth $830B…

Got a question for TIW Office Hours? Hit reply with “Office Hours – [your role]” and your question. I’ll answer a few in future issues — with your name or anonymously, your choice.

⚒️ TOOL SNAPSHOTS

Futuristic tools within AI, no-code, and productivity

• ⚡ MiMo-V2-Flash - Xiaomi's efficient language model for reasoning & coding tasks. Free to use.

• 📤 Sparkle - Ensuring hassle-free, robust email campaigns for business growth. Free option available.

• 🤖 GetProfile - Enhance AI with self-hosted, structured user profiles & memory. Free to use.

• 🌱 Touched Grass - Personalized badges to celebrate your daily achievements. Payment required.

🖼️ AI ART

Examples of great and trending AI art

🤝SUPPORTED BY GLADLY

Can you scale without chaos?

It's peak season, so volume's about to spike. Most teams either hire temps (expensive) or burn out their people (worse). See what smarter teams do: let AI handle predictable volume so your humans stay great.

Check your options

ℹ️ ABOUT US

The Intelligent Worker helps you to be more productive at work with AI, automation, no-code, and other technologies.

We like real, practical, and tangible use-cases and hate hand-wavy, theoretical, and abstract concepts that don’t drive real-world outcomes.

Our mission is to empower individuals, boost their productivity, and future-proof their careers.

We read all your comments - please provide your feedback!