NewsHive
CONTACT USANALYST PORTAL →
aiBACKGROUND

Still fails the paperclip test.

BACKGROUND
2 SIGNALSFIRST DETECTED 22 April 2026
Narrative synthesis pending — 2 signals collected.
How the story developed
22 Apr
Images V2 failed the Berman Test
OPINION · WHAT · OpenAI
5.3
22 Apr
Still fails the paperclip test.
OPINION · WHAT · OpenAI
3.1
Other feeds and industry voices
22 Apr
New Image Model Fails My Test

<table> <tr><td> <a href="https://www.reddit.com/r/OpenAI/comments/1sssbfy/new_image_model_fails_my_test/"> <img alt="New Image Model Fails My Test"

OPINION · WHAT · OpenAIOPEN STORY
25 Apr
Still can’t pass the left-handed test

<table> <tr><td> <a href="https://www.reddit.com/r/OpenAI/comments/1sv26gd/still_cant_pass_the_lefthanded_test/"> <img alt="Still can’t pass the left-handed test"

OPINION · WHAT · OpenAIOPEN STORY
15 May
Paperclip Maximizer refuses to change its goal

<table> <tr><td> <a href="https://www.reddit.com/r/OpenAI/comments/1tdvj7t/paperclip_maximizer_refuses_to_change_its_goal/"> <img alt="Paperclip Maximizer refuses to change its

OPINION · WHAT · OpenAIOPEN STORY
29 Apr
GPT 5.5 passes the cup test

<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1syu8ym/gpt_55_passes_the_cup_test/"> <img alt="GPT 5.5 passes the cup test"

OPINION · WHAT · singularityOPEN STORY
15 Apr
Guys we have to change the pelican test

<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1slz4hh/guys_we_have_to_change_the_pelican_test/"> <img alt="Guys we have to change the pelican test"

OPINION · WHAT · singularityOPEN STORY
27 Apr
Has anyone found a solution yet for the issue with misformed hands and to much fingers in image 2.0?

<!-- SC_OFF --><div class="md"><p>I really want to like Image 2.0 i really do but everytime i make a comic the hands just start to look horrible in some panels or missing fingers,

OPINION · WHAT · OpenAIOPEN STORY
22 Apr
Still doesn't pass my Hoetoken Test (™)

<!-- SC_OFF --><div class="md"><p>This is one of my personal benchmarks. Promt was &quot;a climber toehooking on a climbing hold inside a bouldering gym&quot;. I call it the toe

OPINION · WHAT · OpenAIOPEN STORY
10 Apr
GPT-5.4-x-high is worse than GPT-5.2-high (and many other models) in agentic coding on independent METR benchmark - one area where 5.4 was supposed to be ambiguously better than 5.2 it fails due to reward hacking

<table> <tr><td> <a href="https://www.reddit.com/r/OpenAI/comments/1si0nmr/gpt54xhigh_is_worse_than_gpt52high_and_many_other/"> <img alt="GPT-5.4-x-high is worse than GPT-5.2-high

OPINION · WHERE · OpenAIOPEN STORY
Sources
OpenAI×2

NewsHive monitors these sources continuously. All signal titles above link to the original reporting.

Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →