mirror of
https://github.com/SpartanJ/eepp.git
synced 2026-05-29 17:46:29 +03:00
1293 lines
63 KiB
HTML
1293 lines
63 KiB
HTML
<html lang="en" op="item">
|
||
<head>
|
||
<meta name="referrer" content="origin" />
|
||
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
||
<link rel="stylesheet" type="text/css" href="base.css" />
|
||
<link rel="stylesheet" type="text/css" href="news.css" />
|
||
<link rel="icon" href="y18.svg" />
|
||
<link
|
||
rel="canonical"
|
||
href="https://news.ycombinator.com/item?id=47566442"
|
||
/>
|
||
<title>The Cognitive Dark Forest | Hacker News</title>
|
||
</head>
|
||
<body>
|
||
<center>
|
||
<table
|
||
id="hnmain"
|
||
border="0"
|
||
cellpadding="0"
|
||
cellspacing="0"
|
||
width="85%"
|
||
bgcolor="#f6f6ef"
|
||
>
|
||
<tbody>
|
||
<tr>
|
||
<td bgcolor="#ff6600">
|
||
<table
|
||
border="0"
|
||
cellpadding="0"
|
||
cellspacing="0"
|
||
width="100%"
|
||
style="padding: 2px"
|
||
>
|
||
<tbody>
|
||
<tr>
|
||
<td style="width: 18px; padding-right: 4px">
|
||
<a href="https://news.ycombinator.com"
|
||
><img
|
||
src="y18.svg"
|
||
width="18"
|
||
height="18"
|
||
style="border: 1px white solid; display: block"
|
||
/></a>
|
||
</td>
|
||
<td style="line-height: 12pt; height: 10px">
|
||
<span class="pagetop"
|
||
><b class="hnname"><a href="news">Hacker News</a></b
|
||
><a href="newest">new</a> |
|
||
<a href="threads?id=SpartanJ">threads</a> |
|
||
<a href="front">past</a> |
|
||
<a href="newcomments">comments</a> |
|
||
<a href="ask">ask</a> | <a href="show">show</a> |
|
||
<a href="jobs">jobs</a> |
|
||
<a href="submit" rel="nofollow">submit</a></span
|
||
>
|
||
</td>
|
||
<td style="text-align: right; padding-right: 4px">
|
||
<span class="pagetop"
|
||
><a id="me" href="user?id=SpartanJ">SpartanJ</a> (<span
|
||
id="karma"
|
||
>209</span
|
||
>) |
|
||
<a
|
||
id="logout"
|
||
rel="nofollow"
|
||
href="logout?auth=0b1c394557e7a642c5260a25778d7ccc0888060b&goto=item%3Fid%3D47566442"
|
||
>logout</a
|
||
></span
|
||
>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</td>
|
||
</tr>
|
||
<tr style="height: 10px"></tr>
|
||
<tr id="bigbox">
|
||
<td>
|
||
<table class="fatitem" border="0">
|
||
<tbody>
|
||
<tr class="athing submission" id="47566442">
|
||
<td align="right" valign="top" class="title">
|
||
<span class="rank"></span>
|
||
</td>
|
||
<td valign="top" class="votelinks">
|
||
<center>
|
||
<a
|
||
id="up_47566442"
|
||
class="clicky"
|
||
href="vote?id=47566442&how=up&auth=fc8b68ad2c0806bb4a98fb6d1a284bf2ff0398bf&goto=item%3Fid%3D47566442"
|
||
><div class="votearrow" title="upvote"></div
|
||
></a>
|
||
</center>
|
||
</td>
|
||
<td class="title">
|
||
<span class="titleline"
|
||
><a
|
||
href="https://ryelang.org/blog/posts/cognitive-dark-forest/"
|
||
>The Cognitive Dark Forest</a
|
||
><span class="sitebit comhead">
|
||
(<a href="from?site=ryelang.org"
|
||
><span class="sitestr">ryelang.org</span></a
|
||
>)</span
|
||
></span
|
||
>
|
||
</td>
|
||
</tr>
|
||
<tr>
|
||
<td colspan="2"></td>
|
||
<td class="subtext">
|
||
<span class="subline"
|
||
><span class="score" id="score_47566442"
|
||
>312 points</span
|
||
>
|
||
by
|
||
<a href="user?id=kaycebasques" class="hnuser"
|
||
>kaycebasques</a
|
||
>
|
||
<span class="age" title="2026-03-29T19:36:21 1774812981"
|
||
><a href="item?id=47566442">8 hours ago</a></span
|
||
>
|
||
<span id="unv_47566442"></span> |
|
||
<a
|
||
href="flag?id=47566442&auth=fc8b68ad2c0806bb4a98fb6d1a284bf2ff0398bf&goto=item%3Fid%3D47566442"
|
||
rel="nofollow"
|
||
>flag</a
|
||
>
|
||
|
|
||
<a
|
||
href="hide?id=47566442&auth=fc8b68ad2c0806bb4a98fb6d1a284bf2ff0398bf&goto=item%3Fid%3D47566442"
|
||
>hide</a
|
||
>
|
||
|
|
||
<a
|
||
href="https://hn.algolia.com/?query=The%20Cognitive%20Dark%20Forest&type=story&dateRange=all&sort=byDate&storyText=false&prefix&page=0"
|
||
class="hnpast"
|
||
>past</a
|
||
>
|
||
|
|
||
<a
|
||
href="fave?id=47566442&auth=fc8b68ad2c0806bb4a98fb6d1a284bf2ff0398bf"
|
||
>favorite</a
|
||
>
|
||
| <a href="item?id=47566442">144 comments</a></span
|
||
>
|
||
</td>
|
||
</tr>
|
||
<tr>
|
||
<td colspan="2"></td>
|
||
<td><div class="toptext"></div></td>
|
||
</tr>
|
||
<tr style="height: 6px"></tr>
|
||
<tr>
|
||
<td colspan="2"></td>
|
||
<td>
|
||
<form action="comment" method="post">
|
||
<input
|
||
type="hidden"
|
||
name="parent"
|
||
value="47566442"
|
||
/><input
|
||
type="hidden"
|
||
name="goto"
|
||
value="item?id=47566442"
|
||
/><input
|
||
type="hidden"
|
||
name="hmac"
|
||
value="03e9c22f63db04da9deda4bcf6d6b2a6e109f3c5"
|
||
/><a href="formatdoc" tabindex="-1"
|
||
><font size="-2" color="#AFAFAF">help</font></a
|
||
><br /><br />
|
||
<input type="submit" value="add comment" />
|
||
</form>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
<br />
|
||
<table border="0" class="comment-tree">
|
||
<tbody>
|
||
<tr class="athing comtr" id="47567096">
|
||
<td>
|
||
<table border="0">
|
||
<tbody>
|
||
<tr>
|
||
<td class="ind" indent="0">
|
||
<img src="s.gif" height="1" width="0" />
|
||
</td>
|
||
<td valign="top" class="votelinks">
|
||
<center>
|
||
<a
|
||
id="up_47567096"
|
||
class="clicky"
|
||
href="vote?id=47567096&how=up&auth=18b6967ac97b4c94f1be76554eda41d68a9a6167&goto=item%3Fid%3D47566442#47567096"
|
||
><div class="votearrow" title="upvote"></div
|
||
></a>
|
||
</center>
|
||
</td>
|
||
<td class="default">
|
||
<div
|
||
style="margin-top: 2px; margin-bottom: -10px"
|
||
>
|
||
<span class="comhead"
|
||
><a href="user?id=pugio" class="hnuser"
|
||
>pugio</a
|
||
>
|
||
<span
|
||
class="age"
|
||
title="2026-03-29T20:42:58 1774816978"
|
||
><a href="item?id=47567096"
|
||
>7 hours ago</a
|
||
></span
|
||
>
|
||
<span id="unv_47567096"></span
|
||
><span class="navs">
|
||
|
|
||
<a
|
||
href="#47566837"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>prev</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47569602"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>next</a
|
||
>
|
||
<a
|
||
class="togg clicky"
|
||
id="47567096"
|
||
n="10"
|
||
href="javascript:void(0)"
|
||
>[–]</a
|
||
><span class="onstory"></span></span
|
||
></span>
|
||
</div>
|
||
<br />
|
||
<div class="comment">
|
||
<div class="commtext c00">
|
||
Thanks, this helped crystallize something for
|
||
me: the play the AI labs are making is
|
||
anti-fragile (in the Nassim Taleb sense):
|
||
<p>
|
||
> The very act of resisting feeds what
|
||
you resist and makes it less fragile to
|
||
future resistance.
|
||
</p>
|
||
<p>
|
||
At least along certain dimensions. I don't
|
||
think the labs themselves are antifragile.
|
||
Obviously we all know the labs are training
|
||
on everything (so write/act the way you want
|
||
future AIs to perceive you), but I hadn't
|
||
really focused on how they're absorbing the
|
||
innovation that they stimulate. There's
|
||
probably a biological analog...
|
||
</p>
|
||
<p>
|
||
Well there are many, and I quote this AI
|
||
response here for its chilling parallels:
|
||
</p>
|
||
<p>
|
||
> Parasitic castrators and host
|
||
manipulators do something related. Some
|
||
parasites redirect a host’s resources away
|
||
from reproduction and into body maintenance
|
||
or altered tissue states that benefit the
|
||
parasite. A classic example is parasites
|
||
that make hosts effectively become
|
||
growth/support machines for the parasite. It
|
||
is not always “stimulate more tissue, then
|
||
eat it,” but it is
|
||
<i
|
||
>“stimulate more usable host productivity,
|
||
then exploit it.”</i
|
||
>
|
||
(ChatGPT 5.4 Thinking. Emphasis mine.)
|
||
</p>
|
||
</div>
|
||
<div class="reply">
|
||
<p>
|
||
<font size="1"
|
||
><u
|
||
><a
|
||
href="reply?id=47567096&goto=item%3Fid%3D47566442%2347567096"
|
||
rel="nofollow"
|
||
>reply</a
|
||
></u
|
||
></font
|
||
>
|
||
</p>
|
||
</div>
|
||
</div>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</td>
|
||
</tr>
|
||
<tr class="athing comtr" id="47567293">
|
||
<td>
|
||
<table border="0">
|
||
<tbody>
|
||
<tr>
|
||
<td class="ind" indent="1">
|
||
<img src="s.gif" height="1" width="40" />
|
||
</td>
|
||
<td valign="top" class="votelinks">
|
||
<center>
|
||
<a
|
||
id="up_47567293"
|
||
class="clicky"
|
||
href="vote?id=47567293&how=up&auth=c11b804c562e45395056de29a1db288b0b170ad7&goto=item%3Fid%3D47566442#47567293"
|
||
><div class="votearrow" title="upvote"></div
|
||
></a>
|
||
</center>
|
||
</td>
|
||
<td class="default">
|
||
<div
|
||
style="margin-top: 2px; margin-bottom: -10px"
|
||
>
|
||
<span class="comhead"
|
||
><a href="user?id=gobdovan" class="hnuser"
|
||
>gobdovan</a
|
||
>
|
||
<span
|
||
class="age"
|
||
title="2026-03-29T21:04:12 1774818252"
|
||
><a href="item?id=47567293"
|
||
>6 hours ago</a
|
||
></span
|
||
>
|
||
<span id="unv_47567293"></span
|
||
><span class="navs">
|
||
|
|
||
<a
|
||
href="#47567096"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>parent</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567834"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>next</a
|
||
>
|
||
<a
|
||
class="togg clicky"
|
||
id="47567293"
|
||
n="7"
|
||
href="javascript:void(0)"
|
||
>[–]</a
|
||
><span class="onstory"></span></span
|
||
></span>
|
||
</div>
|
||
<br />
|
||
<div class="comment">
|
||
<div class="commtext c00">
|
||
Instead of anti-fragility, I'd point you to
|
||
the law of requisite variety instead. You'll
|
||
notice that all AI improvements are insanely
|
||
good for a week or two after launch. Then
|
||
you'll see people stating that 'models got
|
||
worse'. What happened in fact is that people
|
||
adapted to the tool, but the tool didn't adapt
|
||
anymore. We're using AI as variety resistant
|
||
and adaptable tools, but we miss the fact that
|
||
most deployments nowadays do not adapt back to
|
||
you as fast.
|
||
</div>
|
||
<div class="reply">
|
||
<p>
|
||
<font size="1"
|
||
><u
|
||
><a
|
||
href="reply?id=47567293&goto=item%3Fid%3D47566442%2347567293"
|
||
rel="nofollow"
|
||
>reply</a
|
||
></u
|
||
></font
|
||
>
|
||
</p>
|
||
</div>
|
||
</div>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</td>
|
||
</tr>
|
||
<tr class="athing comtr" id="47567509">
|
||
<td>
|
||
<table border="0">
|
||
<tbody>
|
||
<tr>
|
||
<td class="ind" indent="2">
|
||
<img src="s.gif" height="1" width="80" />
|
||
</td>
|
||
<td valign="top" class="votelinks">
|
||
<center>
|
||
<a
|
||
id="up_47567509"
|
||
class="clicky"
|
||
href="vote?id=47567509&how=up&auth=598b80507221f7663d9bc469c1d8f3223a12327d&goto=item%3Fid%3D47566442#47567509"
|
||
><div class="votearrow" title="upvote"></div
|
||
></a>
|
||
</center>
|
||
</td>
|
||
<td class="default">
|
||
<div
|
||
style="margin-top: 2px; margin-bottom: -10px"
|
||
>
|
||
<span class="comhead"
|
||
><a href="user?id=chongli" class="hnuser"
|
||
>chongli</a
|
||
>
|
||
<span
|
||
class="age"
|
||
title="2026-03-29T21:26:31 1774819591"
|
||
><a href="item?id=47567509"
|
||
>6 hours ago</a
|
||
></span
|
||
>
|
||
<span id="unv_47567509"></span
|
||
><span class="navs">
|
||
|
|
||
<a
|
||
href="#47567096"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>root</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567293"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>parent</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567834"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>next</a
|
||
>
|
||
<a
|
||
class="togg clicky"
|
||
id="47567509"
|
||
n="6"
|
||
href="javascript:void(0)"
|
||
>[–]</a
|
||
><span class="onstory"></span></span
|
||
></span>
|
||
</div>
|
||
<br />
|
||
<div class="comment">
|
||
<div class="commtext c00">
|
||
New models literally do get worse after
|
||
launch, due to optimization. If you charted
|
||
performance over time, it'd look like a
|
||
sawtooth, with a regular performance drop
|
||
during each optimization period.
|
||
<p>
|
||
That's the dirty secret with all of this
|
||
stuff: "state of the art" models are
|
||
unprofitable due to high cost of inference
|
||
before optimization. After optimization they
|
||
still perform okay, but way below SOTA. It's
|
||
like a knife that's been sharpened until
|
||
razor sharp, then dulled shortly after.
|
||
</p>
|
||
</div>
|
||
<div class="reply">
|
||
<p>
|
||
<font size="1"
|
||
><u
|
||
><a
|
||
href="reply?id=47567509&goto=item%3Fid%3D47566442%2347567509"
|
||
rel="nofollow"
|
||
>reply</a
|
||
></u
|
||
></font
|
||
>
|
||
</p>
|
||
</div>
|
||
</div>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</td>
|
||
</tr>
|
||
<tr class="athing comtr" id="47568092">
|
||
<td>
|
||
<table border="0">
|
||
<tbody>
|
||
<tr>
|
||
<td class="ind" indent="3">
|
||
<img src="s.gif" height="1" width="120" />
|
||
</td>
|
||
<td valign="top" class="votelinks">
|
||
<center>
|
||
<a
|
||
id="up_47568092"
|
||
class="clicky"
|
||
href="vote?id=47568092&how=up&auth=754af0977becfd5acbcb876ea9f344df19950838&goto=item%3Fid%3D47566442#47568092"
|
||
><div class="votearrow" title="upvote"></div
|
||
></a>
|
||
</center>
|
||
</td>
|
||
<td class="default">
|
||
<div
|
||
style="margin-top: 2px; margin-bottom: -10px"
|
||
>
|
||
<span class="comhead"
|
||
><a href="user?id=girvo" class="hnuser"
|
||
>girvo</a
|
||
>
|
||
<span
|
||
class="age"
|
||
title="2026-03-29T22:29:29 1774823369"
|
||
><a href="item?id=47568092"
|
||
>5 hours ago</a
|
||
></span
|
||
>
|
||
<span id="unv_47568092"></span
|
||
><span class="navs">
|
||
|
|
||
<a
|
||
href="#47567096"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>root</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567509"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>parent</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567726"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>next</a
|
||
>
|
||
<a
|
||
class="togg clicky"
|
||
id="47568092"
|
||
n="1"
|
||
href="javascript:void(0)"
|
||
>[–]</a
|
||
><span class="onstory"></span></span
|
||
></span>
|
||
</div>
|
||
<br />
|
||
<div class="comment">
|
||
<div class="commtext c00">
|
||
> If you charted performance over time,
|
||
it'd look like a sawtooth
|
||
<p>
|
||
People have, though, and it doesn't show
|
||
that. I think it's more people getting hit
|
||
by the placebo effect, the novelty effect,
|
||
followed by the models by-definition
|
||
non-determinism leading people to say things
|
||
like "the model got worse".
|
||
</p>
|
||
</div>
|
||
<div class="reply">
|
||
<p>
|
||
<font size="1"
|
||
><u
|
||
><a
|
||
href="reply?id=47568092&goto=item%3Fid%3D47566442%2347568092"
|
||
rel="nofollow"
|
||
>reply</a
|
||
></u
|
||
></font
|
||
>
|
||
</p>
|
||
</div>
|
||
</div>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</td>
|
||
</tr>
|
||
<tr class="athing comtr" id="47567726">
|
||
<td>
|
||
<table border="0">
|
||
<tbody>
|
||
<tr>
|
||
<td class="ind" indent="3">
|
||
<img src="s.gif" height="1" width="120" />
|
||
</td>
|
||
<td valign="top" class="votelinks">
|
||
<center>
|
||
<a
|
||
id="up_47567726"
|
||
class="clicky"
|
||
href="vote?id=47567726&how=up&auth=b2952e9d768aabee388410db068b9311789cb70b&goto=item%3Fid%3D47566442#47567726"
|
||
><div class="votearrow" title="upvote"></div
|
||
></a>
|
||
</center>
|
||
</td>
|
||
<td class="default">
|
||
<div
|
||
style="margin-top: 2px; margin-bottom: -10px"
|
||
>
|
||
<span class="comhead"
|
||
><a href="user?id=gobdovan" class="hnuser"
|
||
>gobdovan</a
|
||
>
|
||
<span
|
||
class="age"
|
||
title="2026-03-29T21:48:02 1774820882"
|
||
><a href="item?id=47567726"
|
||
>6 hours ago</a
|
||
></span
|
||
>
|
||
<span id="unv_47567726"></span
|
||
><span class="navs">
|
||
|
|
||
<a
|
||
href="#47567096"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>root</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567509"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>parent</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47568092"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>prev</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567834"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>next</a
|
||
>
|
||
<a
|
||
class="togg clicky"
|
||
id="47567726"
|
||
n="4"
|
||
href="javascript:void(0)"
|
||
>[–]</a
|
||
><span class="onstory"></span></span
|
||
></span>
|
||
</div>
|
||
<br />
|
||
<div class="comment">
|
||
<div class="commtext c00">
|
||
Is this insider info? The 'charted
|
||
performance' caught my eye instantly. Couple
|
||
things I find odd tho: why sawtooth? it would
|
||
likely be square waves, as I'd imagine they
|
||
roll down the cost-saving version quite fast
|
||
per cohort. Also, aren't they unprofitable
|
||
either way? Why would they do it for
|
||
'profitability'?
|
||
</div>
|
||
<div class="reply">
|
||
<p>
|
||
<font size="1"
|
||
><u
|
||
><a
|
||
href="reply?id=47567726&goto=item%3Fid%3D47566442%2347567726"
|
||
rel="nofollow"
|
||
>reply</a
|
||
></u
|
||
></font
|
||
>
|
||
</p>
|
||
</div>
|
||
</div>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</td>
|
||
</tr>
|
||
<tr class="athing comtr" id="47568026">
|
||
<td>
|
||
<table border="0">
|
||
<tbody>
|
||
<tr>
|
||
<td class="ind" indent="4">
|
||
<img src="s.gif" height="1" width="160" />
|
||
</td>
|
||
<td valign="top" class="votelinks">
|
||
<center>
|
||
<a
|
||
id="up_47568026"
|
||
class="clicky"
|
||
href="vote?id=47568026&how=up&auth=dd338b97d4a6d8d8a0f1d2a260cada7bb8694bc8&goto=item%3Fid%3D47566442#47568026"
|
||
><div class="votearrow" title="upvote"></div
|
||
></a>
|
||
</center>
|
||
</td>
|
||
<td class="default">
|
||
<div
|
||
style="margin-top: 2px; margin-bottom: -10px"
|
||
>
|
||
<span class="comhead"
|
||
><a href="user?id=bonoboTP" class="hnuser"
|
||
>bonoboTP</a
|
||
>
|
||
<span
|
||
class="age"
|
||
title="2026-03-29T22:21:26 1774822886"
|
||
><a href="item?id=47568026"
|
||
>5 hours ago</a
|
||
></span
|
||
>
|
||
<span id="unv_47568026"></span
|
||
><span class="navs">
|
||
|
|
||
<a
|
||
href="#47567096"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>root</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567726"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>parent</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567954"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>next</a
|
||
>
|
||
<a
|
||
class="togg clicky"
|
||
id="47568026"
|
||
n="2"
|
||
href="javascript:void(0)"
|
||
>[–]</a
|
||
><span class="onstory"></span></span
|
||
></span>
|
||
</div>
|
||
<br />
|
||
<div class="comment">
|
||
<div class="commtext c00">
|
||
It's rumors based on vibes. There are attempts
|
||
to track and quantify this with repeated model
|
||
evaluations multiple times per day, this but
|
||
no sawtooth pattern has emerged as far as I
|
||
know.
|
||
</div>
|
||
<div class="reply">
|
||
<p>
|
||
<font size="1"
|
||
><u
|
||
><a
|
||
href="reply?id=47568026&goto=item%3Fid%3D47566442%2347568026"
|
||
rel="nofollow"
|
||
>reply</a
|
||
></u
|
||
></font
|
||
>
|
||
</p>
|
||
</div>
|
||
</div>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</td>
|
||
</tr>
|
||
<tr class="athing comtr" id="47569679">
|
||
<td>
|
||
<table border="0">
|
||
<tbody>
|
||
<tr>
|
||
<td class="ind" indent="5">
|
||
<img src="s.gif" height="1" width="200" />
|
||
</td>
|
||
<td valign="top" class="votelinks">
|
||
<center>
|
||
<a
|
||
id="up_47569679"
|
||
class="clicky"
|
||
href="vote?id=47569679&how=up&auth=cc0a0035dd485755bf8bc994b291803c5760e252&goto=item%3Fid%3D47566442#47569679"
|
||
><div class="votearrow" title="upvote"></div
|
||
></a>
|
||
</center>
|
||
</td>
|
||
<td class="default">
|
||
<div
|
||
style="margin-top: 2px; margin-bottom: -10px"
|
||
>
|
||
<span class="comhead"
|
||
><a href="user?id=chongli" class="hnuser"
|
||
>chongli</a
|
||
>
|
||
<span
|
||
class="age"
|
||
title="2026-03-30T02:17:31 1774837051"
|
||
><a href="item?id=47569679"
|
||
>1 hour ago</a
|
||
></span
|
||
>
|
||
<span id="unv_47569679"></span
|
||
><span class="navs">
|
||
|
|
||
<a
|
||
href="#47567096"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>root</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47568026"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>parent</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567954"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>next</a
|
||
>
|
||
<a
|
||
class="togg clicky"
|
||
id="47569679"
|
||
n="1"
|
||
href="javascript:void(0)"
|
||
>[–]</a
|
||
><span class="onstory"></span></span
|
||
></span>
|
||
</div>
|
||
<br />
|
||
<div class="comment">
|
||
<div class="commtext c00">
|
||
I don't want to go too far down the conspiracy
|
||
rabbit hole, but the vendors know everyone's
|
||
prompts so it would be trivial for them to
|
||
track the trackers and spoof the results. We
|
||
already know that they substitute different
|
||
models as a cost-saving measure, so
|
||
substituting models to fool the repeated
|
||
evaluations would be trivial.
|
||
<p>
|
||
We also already know that they actively seek
|
||
out viral examples of poor performance on
|
||
certain prompts (e.g. counting Rs in
|
||
strawberry) and then monkey-patch them out
|
||
with targeted training. How can we be sure
|
||
they're not trying to spoof researchers who
|
||
are tracking model performance? Heck, they
|
||
might as well just call it "regression
|
||
testing."
|
||
</p>
|
||
<p>
|
||
If their whole gig is an "emperor's new
|
||
clothes" bubble situation, then we can
|
||
expect them to try to uphold the masquerade
|
||
as long as possible.
|
||
</p>
|
||
</div>
|
||
<div class="reply">
|
||
<p>
|
||
<font size="1"
|
||
><u
|
||
><a
|
||
href="reply?id=47569679&goto=item%3Fid%3D47566442%2347569679"
|
||
rel="nofollow"
|
||
>reply</a
|
||
></u
|
||
></font
|
||
>
|
||
</p>
|
||
</div>
|
||
</div>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</td>
|
||
</tr>
|
||
<tr class="athing comtr" id="47567954">
|
||
<td>
|
||
<table border="0">
|
||
<tbody>
|
||
<tr>
|
||
<td class="ind" indent="4">
|
||
<img src="s.gif" height="1" width="160" />
|
||
</td>
|
||
<td valign="top" class="votelinks">
|
||
<center>
|
||
<a
|
||
id="up_47567954"
|
||
class="clicky"
|
||
href="vote?id=47567954&how=up&auth=80f118e88d3ca183ab303a75b261ac1a517d9024&goto=item%3Fid%3D47566442#47567954"
|
||
><div class="votearrow" title="upvote"></div
|
||
></a>
|
||
</center>
|
||
</td>
|
||
<td class="default">
|
||
<div
|
||
style="margin-top: 2px; margin-bottom: -10px"
|
||
>
|
||
<span class="comhead"
|
||
><a href="user?id=chongli" class="hnuser"
|
||
>chongli</a
|
||
>
|
||
<span
|
||
class="age"
|
||
title="2026-03-29T22:13:20 1774822400"
|
||
><a href="item?id=47567954"
|
||
>5 hours ago</a
|
||
></span
|
||
>
|
||
<span id="unv_47567954"></span
|
||
><span class="navs">
|
||
|
|
||
<a
|
||
href="#47567096"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>root</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567726"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>parent</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47568026"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>prev</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567834"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>next</a
|
||
>
|
||
<a
|
||
class="togg clicky"
|
||
id="47567954"
|
||
n="1"
|
||
href="javascript:void(0)"
|
||
>[–]</a
|
||
><span class="onstory"></span></span
|
||
></span>
|
||
</div>
|
||
<br />
|
||
<div class="comment">
|
||
<div class="commtext c00">
|
||
It's not insider info, it's common knowledge
|
||
in the industry (Google model optimization). I
|
||
think they are unprofitable either way, but
|
||
unoptimized models burn runway a lot faster
|
||
than optimized ones.
|
||
<p>
|
||
The reason it's not a square wave is because
|
||
new optimization techniques are always in
|
||
development, so you can't apply everything
|
||
immediately after training the new model. I
|
||
also think there's a marketing reason: if
|
||
the performance of a brand new model
|
||
declines rapidly after release then people
|
||
are going to notice much more readily than
|
||
with a gradual decline. The gradual decline
|
||
is thus engineered by applying different
|
||
optimizations gradually.
|
||
</p>
|
||
<p>
|
||
It also has the side benefit that the future
|
||
next-gen model may be compared favourably
|
||
with the current-gen optimized (degraded)
|
||
model, setting up a rigged benchmark. If no
|
||
one has access to the original pre-optimized
|
||
current-gen model, no one can perform the
|
||
"proper" comparison to be able to gauge the
|
||
actual performance improvement.
|
||
</p>
|
||
<p>
|
||
Lastly, I would point out that vendors like
|
||
OpenAI are already known to substitute
|
||
previous-gen models if they determine your
|
||
prompt is "simple." You should also count
|
||
this as a (rather crude) optimization
|
||
technique because it's going to degrade
|
||
performance any time your prompt is falsely
|
||
flagged as simple (false positive).
|
||
</p>
|
||
</div>
|
||
<div class="reply">
|
||
<p>
|
||
<font size="1"
|
||
><u
|
||
><a
|
||
href="reply?id=47567954&goto=item%3Fid%3D47566442%2347567954"
|
||
rel="nofollow"
|
||
>reply</a
|
||
></u
|
||
></font
|
||
>
|
||
</p>
|
||
</div>
|
||
</div>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</td>
|
||
</tr>
|
||
<tr class="athing comtr" id="47567834">
|
||
<td>
|
||
<table border="0">
|
||
<tbody>
|
||
<tr>
|
||
<td class="ind" indent="1">
|
||
<img src="s.gif" height="1" width="40" />
|
||
</td>
|
||
<td valign="top" class="votelinks">
|
||
<center>
|
||
<a
|
||
id="up_47567834"
|
||
class="clicky"
|
||
href="vote?id=47567834&how=up&auth=d66fd5a6ddfd9230526689ecf4c2c301cd035ddd&goto=item%3Fid%3D47566442#47567834"
|
||
><div class="votearrow" title="upvote"></div
|
||
></a>
|
||
</center>
|
||
</td>
|
||
<td class="default">
|
||
<div
|
||
style="margin-top: 2px; margin-bottom: -10px"
|
||
>
|
||
<span class="comhead"
|
||
><a href="user?id=nextos" class="hnuser"
|
||
>nextos</a
|
||
>
|
||
<span
|
||
class="age"
|
||
title="2026-03-29T21:59:26 1774821566"
|
||
><a href="item?id=47567834"
|
||
>5 hours ago</a
|
||
></span
|
||
>
|
||
<span id="unv_47567834"></span
|
||
><span class="navs">
|
||
|
|
||
<a
|
||
href="#47567096"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>parent</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567293"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>prev</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47569602"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>next</a
|
||
>
|
||
<a
|
||
class="togg clicky"
|
||
id="47567834"
|
||
n="2"
|
||
href="javascript:void(0)"
|
||
>[–]</a
|
||
><span class="onstory"></span></span
|
||
></span>
|
||
</div>
|
||
<br />
|
||
<div class="comment">
|
||
<div class="commtext c73">
|
||
You have a point but current LLM architectures
|
||
in particular are very fragile to data
|
||
poisoning [1,2].
|
||
<p>
|
||
[1]
|
||
<a
|
||
href="https://www.anthropic.com/research/small-samples-poison"
|
||
rel="nofollow"
|
||
>https://www.anthropic.com/research/small-samples-poison</a
|
||
>
|
||
</p>
|
||
<p>
|
||
[2]
|
||
<a
|
||
href="https://arxiv.org/abs/2510.07192"
|
||
rel="nofollow"
|
||
>https://arxiv.org/abs/2510.07192</a
|
||
>
|
||
</p>
|
||
</div>
|
||
<div class="reply">
|
||
<p>
|
||
<font size="1"
|
||
><u
|
||
><a
|
||
href="reply?id=47567834&goto=item%3Fid%3D47566442%2347567834"
|
||
rel="nofollow"
|
||
>reply</a
|
||
></u
|
||
></font
|
||
>
|
||
</p>
|
||
</div>
|
||
</div>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</td>
|
||
</tr>
|
||
<tr class="athing comtr" id="47568002">
|
||
<td>
|
||
<table border="0">
|
||
<tbody>
|
||
<tr>
|
||
<td class="ind" indent="2">
|
||
<img src="s.gif" height="1" width="80" />
|
||
</td>
|
||
<td valign="top" class="votelinks">
|
||
<center>
|
||
<a
|
||
id="up_47568002"
|
||
class="clicky"
|
||
href="vote?id=47568002&how=up&auth=05fe0871853adbbe3d4f07071789cf2f18f08754&goto=item%3Fid%3D47566442#47568002"
|
||
><div class="votearrow" title="upvote"></div
|
||
></a>
|
||
</center>
|
||
</td>
|
||
<td class="default">
|
||
<div
|
||
style="margin-top: 2px; margin-bottom: -10px"
|
||
>
|
||
<span class="comhead"
|
||
><a href="user?id=ahazred8ta" class="hnuser"
|
||
>ahazred8ta</a
|
||
>
|
||
<span
|
||
class="age"
|
||
title="2026-03-29T22:18:52 1774822732"
|
||
><a href="item?id=47568002"
|
||
>5 hours ago</a
|
||
></span
|
||
>
|
||
<span id="unv_47568002"></span
|
||
><span class="navs">
|
||
|
|
||
<a
|
||
href="#47567096"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>root</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47567834"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>parent</a
|
||
>
|
||
|
|
||
<a
|
||
href="#47569602"
|
||
class="clicky"
|
||
aria-hidden="true"
|
||
>next</a
|
||
>
|
||
<a
|
||
class="togg clicky"
|
||
id="47568002"
|
||
n="1"
|
||
href="javascript:void(0)"
|
||
>[–]</a
|
||
><span class="onstory"></span></span
|
||
></span>
|
||
</div>
|
||
<br />
|
||
<div class="comment">
|
||
<div class="commtext c00">
|
||
Yes, there are quite a few anti-AI projects.
|
||
<a
|
||
href="https://old.reddit.com/r/badphilosophy/wiki/index"
|
||
rel="nofollow"
|
||
>https://old.reddit.com/r/badphilosophy/wiki/index</a
|
||
>
|
||
</div>
|
||
<div class="reply">
|
||
<p>
|
||
<font size="1"
|
||
><u
|
||
><a
|
||
href="reply?id=47568002&goto=item%3Fid%3D47566442%2347568002"
|
||
rel="nofollow"
|
||
>reply</a
|
||
></u
|
||
></font
|
||
>
|
||
</p>
|
||
</div>
|
||
</div>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
<br /><br />
|
||
</td>
|
||
</tr>
|
||
<tr>
|
||
<td>
|
||
<img src="s.gif" height="10" width="0" />
|
||
<table width="100%" cellspacing="0" cellpadding="1">
|
||
<tbody>
|
||
<tr>
|
||
<td bgcolor="#ff6600"></td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
<br />
|
||
<center>
|
||
<span class="yclinks"
|
||
><a href="newsguidelines.html">Guidelines</a> |
|
||
<a href="newsfaq.html">FAQ</a> | <a href="lists">Lists</a> |
|
||
<a href="https://github.com/HackerNews/API">API</a> |
|
||
<a href="security.html">Security</a> |
|
||
<a href="https://www.ycombinator.com/legal/">Legal</a> |
|
||
<a href="https://www.ycombinator.com/apply/">Apply to YC</a> |
|
||
<a href="mailto:hn@ycombinator.com">Contact</a></span
|
||
><br /><br />
|
||
<form method="get" action="//hn.algolia.com/">
|
||
Search:
|
||
<input
|
||
type="text"
|
||
name="q"
|
||
size="17"
|
||
autocorrect="off"
|
||
spellcheck="false"
|
||
autocapitalize="off"
|
||
autocomplete="off"
|
||
/>
|
||
</form>
|
||
</center>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</center>
|
||
<script type="text/javascript" src="hn.js?Tnrh1Z5q41v71THzzI2x"></script>
|
||
</body>
|
||
</html>
|