If you're a journalist using AI to write, then yeah, we agree, and I've already said the same thing.

bsky.app/profile/ckd....

FWIW: I *did not* use Gemini to write this thread, but it definitely sounds like I did.

I've not found a single LLM that can write "in my voice" in a way that isn't 1) really cringey and 2) immediately obvious that it was written by an LLM.

I'll do another thread with how I've found it useful..

Other notes:

* The Canvas feature works intermittently at best
* It will not warn you when you exceed the context window. It will just lose its mind and start saying gibberish
* It's annoying/expensive, but it's often worth bouncing between multiple LLMs and comparing the answers.

These 3 things taken together put Gemini ahead of Claude and ChatGPT for me (for now).

IMPORTANT REMINDERS:

* Don't share sensitive information
* Fact check everything it tells you
* Don't take the answers at face value
* Don't have it do your work or write for you
* Remember: you're still smarter

If you ask Gemini to poke holes in your thinking, it will oblige.

If it feels like it's correct and you're wrong, it will stand its ground.

You can wear it down if you want, but then you're using LLMs wrong IMO.

It's not perfect, but it's better than the rest (for now)

Most importantly: Gemini 2.5 Pro is (anecdotally, sample size = me) less sycophantic than Claude or ChatGPT.

It hasn't called me stupid to my face, but there have been a few times where it felt close. And that's a good thing, actually, especially for journalists!

Chain-of-thought by default means it always does a multi-step reasoning process before replying.

This takes a little bit longer, but leads to higher quality answers.

You can also review the reasoning that lead to the response.

This isn't unique to Gemini but Gemini does it really really well.

Larger context window means you can upload more documents, transcripts, notes, etc and the LLM can retain the full content of those documents in it's "memory" during your conversation.

This means you'll get more detailed and accurate answers, and it's great at providing specific citations.

Not sure of the knowledge cutoff but it has real time web search. I did have to fib and say I was only paper trading, because it refuses to cooperate otherwise (reasonably so). (2nd screenshot is from the thinking)

I'll also echo my colleagues in saying that I'll leave the instant it feels like our mission is compromised.

Until then, we will continue investigating and reporting stories like only the Post can - even if that means reporting on ourselves.

and @drewharwell.com wrote a much more comprehensive and reported piece!

www.washingtonpost.com/technology/2...