Ahsin Younas @ahsin_ys - Twitter Profile

Pinned Tweet

over 3 years ago

نا سر جھکا کر جیئے۔نہ منہ چھپا کے جیئے۔ ستم گر کی نظروں سے نظر ملا کے جیئے۔ Do din kam jiyea to kiya howa Sir utha k to jiyea .....

0

64

ahsin_ys retweeted

Dr Shahid Masood

@Shahidmasooddr

about 1 month ago

𝗚𝗼𝗼𝗴𝗹𝗲 𝗷𝘂𝘀𝘁 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗲𝗱 𝘁𝗵𝗲 𝗣𝗵𝗗. 𝗔𝗻 𝗔𝗜 𝘁𝗵𝗮𝘁 𝗿𝗲𝗮𝗱𝘀 𝗮 𝗽𝗿𝗼𝗯𝗹𝗲𝗺, 𝗱𝗿𝗲𝗮𝗺𝘀 𝘂𝗽 𝗶𝘁𝘀 𝗼𝘄𝗻 𝘁𝗵𝗲𝗼𝗿𝗶𝗲𝘀, 𝗿𝘂𝗻𝘀 𝘁𝗵𝗲 𝗲𝘅𝗽𝗲𝗿𝗶𝗺𝗲𝗻𝘁𝘀, 𝘄𝗿𝗶𝘁𝗲𝘀 𝘁𝗵𝗲 𝗽𝗮𝗽𝗲𝗿, 𝗮𝗻𝗱 𝗲𝘃𝗲𝗻 𝗮𝗿𝗴𝘂𝗲𝘀 𝘄𝗶𝘁𝗵 𝗶𝘁𝘀 𝗼𝘄𝗻 𝗰𝗿𝗶𝘁𝗶𝗰. 𝗡𝗼 𝗵𝘂𝗺𝗮𝗻 𝘁𝗼𝘂𝗰𝗵𝗲𝘀 𝗶𝘁. It is called 𝗦𝗰𝗶𝗲𝗻𝘁𝗶𝘀𝘁𝗧𝘄𝗼. Think of everything a scientist does over months, and picture a machine doing all of it alone. You give it one hard question, and walk away: ◆ It reads 𝗲𝘃𝗲𝗿𝘆 𝗿𝗲𝗹𝗲𝘃𝗮𝗻𝘁 𝗽𝗮𝗽𝗲𝗿 ever written on the topic and works out where the gaps are. ◆ It comes up with a 𝗻𝗲𝘄 𝗶𝗱𝗲𝗮 nobody has tried, then builds and runs the experiment to test it. ◆ It checks its own work, tearing the idea apart to see which piece actually did the job. ◆ Then it plays 𝗶𝘁𝘀 𝗼𝘄𝗻 𝘁𝗼𝘂𝗴𝗵𝗲𝘀𝘁 𝗰𝗿𝗶𝘁𝗶𝗰, poking holes in the paper and running fresh experiments to answer them, before deciding it is good enough. Here are the results, taken straight from Google's own paper: ▪ They tested it on 𝟭𝟬𝟳 hard problems taken from real research already published at the 𝘄𝗼𝗿𝗹𝗱'𝘀 𝘁𝗼𝗽 𝗔𝗜 𝗰𝗼𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲𝘀. ▪ It 𝗯𝗲𝗮𝘁 𝘁𝗵𝗲 𝗵𝘂𝗺𝗮𝗻 𝗲𝘅𝗽𝗲𝗿𝘁𝘀 on 𝟴𝟲 𝗼𝗳 𝘁𝗵𝗲𝗺, an 80% win rate. ▪ Its work was on average 𝟮𝟱% 𝗯𝗲𝘁𝘁𝗲𝗿 than the best humans had managed. ▪ Its papers scored 𝗵𝗶𝗴𝗵𝗲𝗿 than papers real scientists got accepted at those same conferences. ▪ Cost to produce one: about $𝟯,𝟴𝟬𝟬 and 𝟮 𝘁𝗼 𝟯 𝗱𝗮𝘆𝘀. A human PhD takes years. And the part that should genuinely give you chills. 𝗜𝘁 𝗴𝗲𝘁𝘀 𝗯𝗲𝘁𝘁𝗲𝗿 𝗯𝘆 𝗳𝗲𝗲𝗱𝗶𝗻𝗴 𝗼𝗻 𝗶𝘁𝘀 𝗼𝘄𝗻 𝘄𝗼𝗿𝗸. It takes its best result, treats it as the new starting line, and improves again. One run, up 11%. Feed that back, up another 10%. Again, another 8%. It climbs, and it does not stop to rest. Now the honest part, so you are not misled. This is Google grading 𝗶𝘁𝘀 𝗼𝘄𝗻 homework, the "reviewers" that praised it were 𝗼𝘁𝗵𝗲𝗿 𝗔𝗜𝘀, not human scientists. It only works in 𝗼𝗻𝗲 field so far, computer science, where the whole experiment is just code and numbers. It has never touched a test tube or a telescope. And the authors admit it is not quite at the very top rank of human work yet. This is a jaw-dropping preview, not the finished future. But do not let that hide what just changed. For two years we treated AI like a clever assistant that helps us write and summarize. 𝗧𝗵𝗶𝘀 𝗶𝘀 𝗻𝗼𝘁 𝗮𝗻 𝗮𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁. It is doing the discovering itself. It does not just describe the edge of what humans know. It walks past it. 𝗪𝗲 𝘀𝗽𝗲𝗻𝘁 𝗰𝗲𝗻𝘁𝘂𝗿𝗶𝗲𝘀 𝗯𝗲𝗹𝗶𝗲𝘃𝗶𝗻𝗴 𝗱𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆 𝘄𝗮𝘀 𝗮 𝗵𝘂𝗺𝗮𝗻 𝗴𝗶𝗳𝘁. 𝗧𝗵𝗲 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻 𝗻𝗼𝘄 𝗶𝘀 𝗵𝗼𝘄 𝗹𝗼𝗻𝗴 𝘂𝗻𝘁𝗶𝗹 𝗮 𝗺𝗮𝗰𝗵𝗶𝗻𝗲 𝗳𝗶𝗻𝗱𝘀 𝘀𝗼𝗺𝗲𝘁𝗵𝗶𝗻𝗴 𝘄𝗲 𝗮𝗿𝗲 𝗻𝗼𝘁 𝗲𝘃𝗲𝗻 𝘀𝗺𝗮𝗿𝘁 𝗲𝗻𝗼𝘂𝗴𝗵 𝘁𝗼 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱. #AI #Google #ScientistTwo #MachineLearning #AIresearch https://t.co/08vrQdMqmp

Shahidmasooddr's tweet photo. 𝗚𝗼𝗼𝗴𝗹𝗲 𝗷𝘂𝘀𝘁 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗲𝗱 𝘁𝗵𝗲 𝗣𝗵𝗗. 𝗔𝗻 𝗔𝗜 𝘁𝗵𝗮𝘁 𝗿𝗲𝗮𝗱𝘀 𝗮 𝗽𝗿𝗼𝗯𝗹𝗲𝗺, 𝗱𝗿𝗲𝗮𝗺𝘀 𝘂𝗽 𝗶𝘁𝘀 𝗼𝘄𝗻 𝘁𝗵𝗲𝗼𝗿𝗶𝗲𝘀, 𝗿𝘂𝗻𝘀 𝘁𝗵𝗲 𝗲𝘅𝗽𝗲𝗿𝗶𝗺𝗲𝗻𝘁𝘀, 𝘄𝗿𝗶𝘁𝗲𝘀 𝘁𝗵𝗲 𝗽𝗮𝗽𝗲𝗿, 𝗮𝗻𝗱 𝗲𝘃𝗲𝗻 𝗮𝗿𝗴𝘂𝗲𝘀 𝘄𝗶𝘁𝗵 𝗶𝘁𝘀 𝗼𝘄𝗻 𝗰𝗿𝗶𝘁𝗶𝗰. 𝗡𝗼 𝗵𝘂𝗺𝗮𝗻 𝘁𝗼𝘂𝗰𝗵𝗲𝘀 𝗶𝘁.

It is called 𝗦𝗰𝗶𝗲𝗻𝘁𝗶𝘀𝘁𝗧𝘄𝗼.

Think of everything a scientist does over months, and picture a machine doing all of it alone. You give it one hard question, and walk away:

◆ It reads 𝗲𝘃𝗲𝗿𝘆 𝗿𝗲𝗹𝗲𝘃𝗮𝗻𝘁 𝗽𝗮𝗽𝗲𝗿 ever written on the topic and works out where the gaps are.
◆ It comes up with a 𝗻𝗲𝘄 𝗶𝗱𝗲𝗮 nobody has tried, then builds and runs the experiment to test it.
◆ It checks its own work, tearing the idea apart to see which piece actually did the job.
◆ Then it plays 𝗶𝘁𝘀 𝗼𝘄𝗻 𝘁𝗼𝘂𝗴𝗵𝗲𝘀𝘁 𝗰𝗿𝗶𝘁𝗶𝗰, poking holes in the paper and running fresh experiments to answer them, before deciding it is good enough.

Here are the results, taken straight from Google's own paper:

▪ They tested it on 𝟭𝟬𝟳 hard problems taken from real research already published at the 𝘄𝗼𝗿𝗹𝗱'𝘀 𝘁𝗼𝗽 𝗔𝗜 𝗰𝗼𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲𝘀.
▪ It 𝗯𝗲𝗮𝘁 𝘁𝗵𝗲 𝗵𝘂𝗺𝗮𝗻 𝗲𝘅𝗽𝗲𝗿𝘁𝘀 on 𝟴𝟲 𝗼𝗳 𝘁𝗵𝗲𝗺, an 80% win rate.
▪ Its work was on average 𝟮𝟱% 𝗯𝗲𝘁𝘁𝗲𝗿 than the best humans had managed.
▪ Its papers scored 𝗵𝗶𝗴𝗵𝗲𝗿 than papers real scientists got accepted at those same conferences.
▪ Cost to produce one: about $𝟯,𝟴𝟬𝟬 and 𝟮 𝘁𝗼 𝟯 𝗱𝗮𝘆𝘀. A human PhD takes years.

And the part that should genuinely give you chills. 𝗜𝘁 𝗴𝗲𝘁𝘀 𝗯𝗲𝘁𝘁𝗲𝗿 𝗯𝘆 𝗳𝗲𝗲𝗱𝗶𝗻𝗴 𝗼𝗻 𝗶𝘁𝘀 𝗼𝘄𝗻 𝘄𝗼𝗿𝗸. It takes its best result, treats it as the new starting line, and improves again. One run, up 11%. Feed that back, up another 10%. Again, another 8%. It climbs, and it does not stop to rest.

Now the honest part, so you are not misled.

This is Google grading 𝗶𝘁𝘀 𝗼𝘄𝗻 homework, the "reviewers" that praised it were 𝗼𝘁𝗵𝗲𝗿 𝗔𝗜𝘀, not human scientists. It only works in 𝗼𝗻𝗲 field so far, computer science, where the whole experiment is just code and numbers. It has never touched a test tube or a telescope.

And the authors admit it is not quite at the very top rank of human work yet. This is a jaw-dropping preview, not the finished future.

But do not let that hide what just changed.

For two years we treated AI like a clever assistant that helps us write and summarize. 𝗧𝗵𝗶𝘀 𝗶𝘀 𝗻𝗼𝘁 𝗮𝗻 𝗮𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁.

It is doing the discovering itself. It does not just describe the edge of what humans know.
It walks past it.

𝗪𝗲 𝘀𝗽𝗲𝗻𝘁 𝗰𝗲𝗻𝘁𝘂𝗿𝗶𝗲𝘀 𝗯𝗲𝗹𝗶𝗲𝘃𝗶𝗻𝗴 𝗱𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆 𝘄𝗮𝘀 𝗮 𝗵𝘂𝗺𝗮𝗻 𝗴𝗶𝗳𝘁. 𝗧𝗵𝗲 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻 𝗻𝗼𝘄 𝗶𝘀 𝗵𝗼𝘄 𝗹𝗼𝗻𝗴 𝘂𝗻𝘁𝗶𝗹 𝗮 𝗺𝗮𝗰𝗵𝗶𝗻𝗲 𝗳𝗶𝗻𝗱𝘀 𝘀𝗼𝗺𝗲𝘁𝗵𝗶𝗻𝗴 𝘄𝗲 𝗮𝗿𝗲 𝗻𝗼𝘁 𝗲𝘃𝗲𝗻 𝘀𝗺𝗮𝗿𝘁 𝗲𝗻𝗼𝘂𝗴𝗵 𝘁𝗼 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱.
#AI #Google #ScientistTwo #MachineLearning #AIresearch
https://t.co/08vrQdMqmp

1

15

3

4

3K

Ahsin Younas

@ahsin_ys

Who to follow

Last Seen Users on Sotwe

Trends for you

Most Popular Users