At the start of 2025, I made some optimistic predictions about where AI would go this year. Now that we’re wrapping up December, let’s see how they held up.

My Original Predictions

Here’s what I posted on LinkedIn in January 2025:

My optimistic predictions for #AI in 2025:

• 💻 More local models: models integrated into browsers and operating systems
• 💡 Increase in adoption of domain specific Small Language Models (SLM)
• 📊 More affordable fine-tuning and training
• ⌨️ Companies will hire specific AI agents
• 🎬 80% of a movie shall be baked with AI
• 🏥 AI-assisted medical diagnosis and surgical interventions save lives
• 🚫 Lawsuits against AI and newer government policies

Verdict: 6.5/7 correct. Not bad for optimism.

The Scorecard

Prediction	Status	Reality Check
💻 More local models in browsers/OS	✅	Chrome, Edge, Safari, Brave, Opera, Arc all ship AI
💡 Domain-specific SLMs take off	✅	SLMs outperform LLMs in niche tasks
📊 Affordable fine-tuning	✅	Fine-tuning now $5k-$50k range
⌨️ Companies hire AI agents	✅	62% experimenting, 23% scaling
🎬 80% of a movie baked with AI	⚠️	Indie yes, Hollywood not quite
🏥 AI-assisted diagnosis saves lives	✅	66% of physicians using AI
🚫 Lawsuits & gov policies	✅	Dec 11 Executive Order

Deep Dive

💻 Local Models Everywhere

The “edge AI” revolution happened. Every major browser now has AI built-in:

Browser	AI Integration
Chrome	Gemini + window.ai API
Edge	Copilot (OpenAI)
Safari	Apple Intelligence
Brave	Leo (Llama 3, Mixtral, optional GPT-4o)
Opera	Aria (Opera’s model + OpenAI)
Arc	Arc Max (GPT-4 class models)
Firefox	Add-ons only (ChatGPT, Gemini, Claude, Local)
Vivaldi	External AI tools integration

Plus standalone tools:

• Jan AI - runs LLMs completely offline
• Perplexity Comet - AI-first browser (launched July 2025)
• WebLLM/WebGPU - bringing models to the browser

💡 Small Language Models Win

The market shifted from “bigger is better” to “right-sized for the job”:

• Companies like AT&T, EY, Thomson Reuters embraced SLMs
• Training costs 1/10th of LLMs
• NVIDIA research shows SLMs effective as autonomous agents
• Microsoft Phi-3 driving enterprise adoption

📊 Fine-Tuning for Everyone

Training frontier models still costs $100M+, but:

• Fine-tuning dropped to $5k-$50k range
• GenAI MVP now costs $50k-$150k
• OpenAI fine-tuning: GPT-4.1 nano at $1.50/1M tokens
• Custom AI no longer enterprise-only

⌨️ The Agent Economy

2025 was the “Year of Agents”:

• Companies subscribing to AI agents for specific tasks
• 62% of organizations experimenting with AI agents
• 23% scaling agentic AI in production
• Agents handling customer support, coding, research autonomously

🎬 AI Cinema (Partial Hit)

The “80%” mark happened in indie, not Hollywood:

• Daisy - AI horror short by Storybook Studios
• Horizon 2080 - 57-minute sci-fi epic by Epic AI Films
• Hollywood uses AI heavily for VFX (~30-50%)
• Full AI features remain indie territory

🏥 AI Saves Lives

AI-assisted diagnosis became standard:

• 66% of physicians using AI (up from 38% in 2023)
• AMA launched Center for Digital Health and AI
• 90% of hospitals use AI for early diagnosis
• AI imaging detects anomalies faster than radiologists

🚫 The Legal Reckoning

On December 11, 2025:

• Executive Order established DOJ “AI Litigation Task Force”
• Challenges state AI laws conflicting with federal policy
• California’s strict regulations vs. federal frameworks
• Exactly the friction predicted

What’s Next for 2026?

Stay tuned for my 2026 predictions. Spoiler: agents get autonomous, regulation gets serious, and the line between AI-generated and human-created content gets very blurry.

---

Original LinkedIn post from January 2025.