Search
Search titles only
By:
Search titles only
By:
Log in
Register
Search
Search titles only
By:
Search titles only
By:
Menu
Install the app
Install
Forums
New posts
All threads
Latest threads
New posts
Trending threads
Trending
Search forums
What's new
New posts
New ads
New profile posts
Latest activity
Free Ads
Latest reviews
Search ads
Members
Current visitors
New profile posts
Search profile posts
Contact us
Latest ads
Ad icon
Iptv
musicking
Updated:
Yesterday at 9:52 AM
Ad icon
ZTE MF283U 4G Unlocked Router (Used)
ayanthamaxi
Updated:
Jul 19, 2026
ලංකාවේ හොඳම උපකාරක පන්ති සහ ගුරුවරුන් එකම තැනකින් - TopTuition.lk
dulithapathum
Updated:
Jul 18, 2026
Colombo
RidhMathraa ’26 🎶✨
Tmadhusanka
Updated:
Jul 15, 2026
Ad icon
Colombo
PXN V10 Pro Direct Drive Racing Wheel (Under Warranty)
Abdur Rahman
Updated:
Jul 15, 2026
Electronics
Vehicles
Property
Search
Reply to thread
Forums
General
ElaKiri Talk!
How many Rs are there in Strawberry? Claude Sonnet 3.5 vs ChatGPT 4o
Get the App
JavaScript is disabled. For a better experience, please enable JavaScript in your browser before proceeding.
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an
alternative browser
.
Message
<blockquote data-quote="AiLankan" data-source="post: 30259329" data-attributes="member: 586885"><p>Instruction tuning doesn't work like that. One thing lacking in LLMs is called reasoning. But the community is catching up! Best example is open AI O1.</p><p></p><p></p><p>> උදාහරණයක් විදිහට අපි illegal දෙයක් අැහුවොත් එවෙලෙම එ්කට restrictions දාන්න instructions කොහොමත් තියෙනවනෙ. ඉතින් මේ වගේ සුළු දෙයක් බැරිද? කොටින්ම add spaces between each letter of the word කියල extra prompt එකක් දැම්මා නම්, දෙවිහෙන් එන උත්තර වෙනස්ද කියල බලලා වෙනස් නම් space කඩපු එක වඩාත් හරි වීමේ probability එක වැඩි නිසා එ්ක උත්තරය විදිහට දෙන්න හදල නැත්තෙ අැයි?</p><p></p><p>Usually, these kinds of reasoning questions are helpful for scientists to improve their systems, so it is a stupid thing to hardcode a solution. Even my research team tries hard to improve LLMs to reason. </p><p> Another thing is that open AI doesn't necessarily put restrictions on everything, and models are capable of not answering such questions. You can read about "Reinforcement Learning from Human Feedback"</p></blockquote><p></p>
[QUOTE="AiLankan, post: 30259329, member: 586885"] Instruction tuning doesn't work like that. One thing lacking in LLMs is called reasoning. But the community is catching up! Best example is open AI O1. > උදාහරණයක් විදිහට අපි illegal දෙයක් අැහුවොත් එවෙලෙම එ්කට restrictions දාන්න instructions කොහොමත් තියෙනවනෙ. ඉතින් මේ වගේ සුළු දෙයක් බැරිද? කොටින්ම add spaces between each letter of the word කියල extra prompt එකක් දැම්මා නම්, දෙවිහෙන් එන උත්තර වෙනස්ද කියල බලලා වෙනස් නම් space කඩපු එක වඩාත් හරි වීමේ probability එක වැඩි නිසා එ්ක උත්තරය විදිහට දෙන්න හදල නැත්තෙ අැයි? Usually, these kinds of reasoning questions are helpful for scientists to improve their systems, so it is a stupid thing to hardcode a solution. Even my research team tries hard to improve LLMs to reason. Another thing is that open AI doesn't necessarily put restrictions on everything, and models are capable of not answering such questions. You can read about "Reinforcement Learning from Human Feedback" [/QUOTE]
Insert quotes…
Verification
Hath warak paha keeyada? (hatha wadikireema paha)
Post reply
Top
Bottom