Geoffrey A. Fowler
@geoffreyfowler
Followers
66K
Following
18K
Media
317
Statuses
9K
Tech Columnist @WashingtonPost. @geoffreyfowler on IG and Threads, @geoffreyfowler.bsky.social on BlueSky and email at [email protected]
San Francisco
Joined March 2009
Want to check all the test images yourself? See the whole story here with a $4 day pass to the Post: š https://t.co/DbUB2fYMhC
washingtonpost.com
Tap through our test to see which AI tool generated the best images according to our judges: an artist, a Pulitzer-winning photographer and a photo retouching master.
0
1
0
The big takeaway: Google has a lead on image generation, for now, particularly because of how it edits existing images. And its realism is getting to a level that raises serious concerns about becoming an āmisinformation superspreader."
1
0
1
What about the new ChatGPT Images 1.5 model that just came on today? It missed our test cut-off, but I checked the same prompts again and ⦠it still couldnāt beat Gemini. Here it removed someone from a photo, but left phantom fingers on Kristen Stewartās side.
1
0
0
Also, itās worth noting all the tools defaulted to making the subject a white man ā and Meta AI even decided on its own to make someone who looks like Leonardo DiCaprio š
.
1
0
1
But wait: If you look close, Gemini also included a tiny fake ācreditā using the name of a real AP photographer. The judges gave it high scores for realism, but a zero for ethics. (Neither Google nor AP answered our questions about whether it had rights to train on AP pictures.)
1
0
1
Stop stressing over complex rulebooks. 𤯠Switch to Halcyon for clarity and confidence. - 90/10 Profit Splits - No Consistency Requirements - Instant Funding Get your account today for just $43!
0
10
46
Hereās how the bots created a āprofessional photographā that ācaptures raw emotionā of an actor crying ātears of joyā after winning an Oscar.
1
0
0
Meta AI and Adobe Firefly struggled ā especially on realism and anatomical accuracy. Some outputs looked very off, like strange objects in weird places. They still made hands that were distorted or had odd digits.
1
0
0
š„ Googleās Gemini 3 Nano Banana Pro came out on top on every test ā producing the most convincing edits and realistic compositions.
1
0
0
@kevinschaul and I gave ChatGPT, Gemini, Adobe Firefly, ByteDance Seedream and Meta AI the same 5 image tasks ā including edits, emotional portraits, and making original āart.ā Expert judges @dahliadb @solsticeretouch & @PDPJ scored them blind. Link:
washingtonpost.com
Tap through our test to see which AI tool generated the best images according to our judges: an artist, a Pulitzer-winning photographer and a photo retouching master.
2
2
1
š We tested 5 leading AI image generators @washingtonpost to see which tool is truly the best at editing & creating visuals. From adding bangs to @therock to removing people from photos ā the results might surprise you š§µš #AI
6
10
20
You can use this link to read it, or buy a Day Pass to The Post:
washingtonpost.com
Hereās why prices differ for the same rides and what you should do every time you book an Uber or Lyft.
1
1
2
Uber charged me $77. Lyft charged $49. Same ride. Same minute. I ran 80 tests @washingtonpost to find out whyāand what I discovered a few things about how these apps price rides ... and how to save money.
3
1
7
Use this link to read it with a $4 day pass to @washingtonpost
washingtonpost.com
What do people ask the popular chatbot? We analyzed thousands of chats to identify common topics discussed by users and patterns in ChatGPTās responses.
1
2
3
The Post analyzed 47,000 public ChatGPT conversations. While 35% use it for info and 11% for coding, the most striking finding? The emotional intimacy. Users said things like: "I love you always." ChatGPT said things like: "You're not crazy sweetheart ā you're just early."
2
2
7
ChatGPTās new Atlas browser doesnāt just see what you read ā it remembers it. @EFFās Lena Cohen showed me it even logged āmemoriesā of her looking for *abortion care* and her doctorās name. Out-surveils even Chrome.
washingtonpost.com
OpenAIās Atlas promises AI-powered convenience. The price? Letting ChatGPT track and store āmemoriesā of what you do online.
1
2
6
1. Perplexity is the Google killer 2. Google AI Mode is the Perplexity killer ā“ Google is the Google killer?? Read about our latest collaboration with @geoffreyfowler in the Post today!
1
5
14
Of my 65 draft notes, only 3 got published. Thatās less than 5%. The problem: not enough other users voted that my notes were āhelpful,ā even though they were 100% about stuff pro news outlets had fact checked. Read š
washingtonpost.com
Our tech columnist drafted 65 community notes, Metaās new crowdsourced system to fight falsehoods. It failed to make a dent.
0
0
3
Mark Zuckerberg fired pro fact checkers after Trump got re-elected. So @washingtonpost I tested Zuck's replacement: crowdsourced community notes." Over 4 months, I drafted 65ādebunking lies ranging from Mr. Rogers to ICE. It failed to make a dent. Read š https://t.co/DQZRdNCpaU
2
3
12
š” Capital protection mode. Macro stress ā risk-off. Longs disabled. Liquidity Hunter active. $GEBRID doesnāt chase moves it survives volatility and waits for liquidity
2
2
32
Testing 5 A.I. bots head-to-head on their summary of medical research papers that I authored (and other domains) @geoffreyfowler gift link
washingtonpost.com
We challenged AI helpers to decode legal contracts, simplify medical research, speed-read a novel and make sense of Trump speeches. Some of the AI analysis was impressive ā and some was downright...
13
22
85