AI Tidbits – Search, Creativity, Trust: Where AI Stands as 2025 Ends

AI Tidbits – Search, Creativity, Trust: Where AI Stands as 2025 Ends

If 2025 taught us anything, it’s this: AI didn’t quietly integrate into our lives — it moved straight into the center of attention. Search, creativity, commerce, and even humor are now AI-shaped. As we close the year, this, slightly long, final edition of AI Tidbits captures where the AI world stands today — powerful, useful, controversial, and occasionally hilarious.

 

📢 ChatGPT Starts Testing Sponsored Answers

OpenAI has officially started testing sponsored responses inside ChatGPT conversations. Unlike traditional banner ads, these appear inline — woven directly into answers, clearly labelled but part of the conversational flow.

This marks a major strategic shift. ChatGPT is beginning to resemble a search engine with intent-driven monetisation, driven by user questions rather than page views. With OpenAI facing massive compute costs, this move feels inevitable — but it raises an important question.

Just like Google years ago, ChatGPT now sits at the crossroads of trust vs scale. How OpenAI balances relevance, transparency, and user confidence will decide whether AI remains a trusted guide… or slowly turns into a marketplace.

Interesting times ahead?

 

📌 Pinned Chats in ChatGPT

A small update, but a very practical one.

You can now pin important chats to the top of your ChatGPT list for quick access — across iOS, Android, Web, and Atlas.

• On web: hover over a chat → click ⋯ → Pin chat
• On mobile: long-press a chat → Pin chat

For anyone running long projects, this instantly reduces friction and clutter.

 

📸 Product Photography Tip

If you’re creating premium-looking product visuals, this prompt delivers consistently strong results.

Prompt:
[A PRODUCT NAME] on the cover, set against an elegant and luxurious [BACKGROUND STYLE / COLORS] with [LIGHTING DESCRIPTION] radiating from behind it.
The product name “[PRODUCT NAME]” appears at the top center in [FONT STYLE / COLOR], with the [BRAND LOGO / SYMBOL] placed just below. Additional text highlights key product details reading: “[FEATURE / SHADE / BENEFIT LINE].” A subtle call-to-action label sits beneath the description. The [PRODUCT PACKAGING DETAILS] feature [METALLIC / TEXTURED / PREMIUM DESIGN ELEMENTS], giving it a high-end, luxurious finish

Perfect for ads, ecommerce listings, or pitch decks.

 

🖼️ Restore Old Photos with Nano Banana

One of the most emotionally satisfying uses of AI.

Go to Tools → Create Image 🍌, upload your old or damaged photo, and use this:

Prompt:
Faithfully restore this image with high fidelity to modern photograph quality, in full color, upscale to 4K

It’s remarkable how well AI can now revive memories without over-stylising them.

 

🤖 The Year’s Most Asked Question: Which LLM Should You Use?

Answered simply by Charlie Hills, based on real usage — not hype.

• One subscription only?
  Go with Perplexity Pro. You get solid access to GPT-5 and Claude, plus arguably the best AI-powered web search experience available.
• Developer focused?
  Claude Sonnet 4.5 stands out for coding, especially with its Projects feature for managing complex codebases.
• Need deep reasoning + multimodal (audio/video)?
  Gemini 3 Pro currently leads, with top reasoning scores and an excellent context window.

Charlie now relies almost entirely on Gemini 3 Pro — something he never expected to say.
Honestly? Neither did I.

If you want to read the details, here they are:

 

1. Gemini 3 Pro — 71/80		2. Claude Sonnet 4.5 — 63/80
Best reasoning model available. First to break 1500 on LMArena leaderboard. Wins most benchmark tests. Handles text, images, video, audio together. Massive 1M token context window.		World’s best coding model. Fixes real GitHub bugs better than any competitor. Runs autonomous tasks for 30+ hours straight. Zero errors on code editing tests.
Coding: █████████░ 9/10		Coding: ██████████ 10/10
Reasoning: ██████████ 10/10		Reasoning: █████████░ 9/10
Math: █████████░ 9/10		Math: ███████░░░ 7/10
Speed: █████████░ 9/10		Speed: ███████░░░ 7/10
Cost: ███████░░░ 7/10		Cost: █████░░░░░ 5/10
Context: ██████████ 10/10		Context: ███████░░░ 7/10
Web Search: █████████░ 9/10		Web Search: ███░░░░░░░ 3/10
Ecosystem: ████████░░ 8/10		Ecosystem: ████████░░ 8/10

 

3. GPT-5 — 63/80		4. Perplexity Pro — 58/80
Best developer tools and integrations. Automatically switches between fast mode and thinking mode. Biggest ecosystem with most third-party support. Works everywhere.		One subscription gets you GPT-5, Claude, Gemini and more. Best web search with live citations. Perfect for research. No need to pick models yourself.
Coding: ██████████ 10/10		Coding: ████████░░ 8/10
Reasoning: ██████████ 10/10		Reasoning: ████████░░ 8/10
Math: █████████░ 9/10		Math: ████████░░ 8/10
Speed: ████████░░ 8/10		Speed: ███████░░░ 7/10
Cost: ████░░░░░░ 4/10		Cost: ████░░░░░░ 4/10
Context: ██████░░░░ 6/10		Context: ███████░░░ 7/10
Web Search: ██████░░░░ 6/10		Web Search: ██████████ 10/10
Ecosystem: ██████████ 10/10		Ecosystem: ██████░░░░ 6/10

 

5. Grok 4.1 — 55/80		6. DeepSeek V3.2 — 51/80
Most human-like conversations. Ranks #1 for personality and creativity. Plugged into X for real-time info. Reduced mistakes by 66%. Best creative writing.		Destroyed math competitions. Gold medals at IMO, IOI, ICPC, CMO. Beats GPT-5 at pure math. 10x cheaper than competitors. Open source and free to modify.
Coding: ████████░░ 8/10		Coding: █████████░ 9/10
Reasoning: ███████░░░ 7/10		Reasoning: █████████░ 9/10
Math: ███████░░░ 7/10		Math: ██████████ 10/10
Speed: ████████░░ 8/10		Speed: ███░░░░░░░ 3/10
Cost: ██████░░░░ 6/10		Cost: ██████████ 10/10
Context: █████░░░░░ 5/10		Context: █████░░░░░ 5/10
Web Search: █████████░ 9/10		Web Search: █░░░░░░░░░ 1/10
Ecosystem: █████░░░░░ 5/10		Ecosystem: ████░░░░░░ 4/10

 

7. Copilot — 49/80		8. Meta AI — 62/80
GPT-5 but slower and more restricted. Needs Microsoft 365 for best features. Only searches your OneDrive files. Good for enterprises already using Microsoft.		Llama 4 powers Facebook, Instagram, WhatsApp. Handles 1M tokens at once. Beats GPT-4o on most tests. Open source means you can customise everything.
Coding: ████████░░ 8/10		Coding: ████████░░ 8/10
Reasoning: ████████░░ 8/10		Reasoning: ████████░░ 8/10
Math: ████████░░ 8/10		Math: ████████░░ 8/10
Speed: ██████░░░░ 6/10		Speed: ████████░░ 8/10
Cost: ███░░░░░░░ 3/10		Cost: █████████░ 9/10
Context: █████░░░░░ 5/10		Context: ██████████ 10/10
Web Search: █████░░░░░ 5/10		Web Search: ████░░░░░░ 4/10
Ecosystem: ██████░░░░ 6/10		Ecosystem: ███████░░░ 7/10

 

The takeaway: Stop stacking subscriptions you don’t fully use. Pick the tool that matches your real needs and build your AI stack intentionally.

 

😂 To Close the Year: LOL

If AI can make fun of itself — and do it well — that says a lot about where we are.

🎉 Happy New Year 2026

Thank you for reading AI Tidbits all year — for questioning, experimenting, and staying curious while the ground kept shifting beneath us.

Enjoy the celebrations, recharge, and get ready.
2026 is not going to slow down. 🚀✨