David Siska ๐ŸŒ

985 posts

David Siska ๐ŸŒ banner
David Siska ๐ŸŒ

David Siska ๐ŸŒ

@dsiska

Stochastic control, RL, Maths of AI; Reader at School of Mathematics, University of Edinburgh @EdinUniMaths

Edinburgh, Scotland ๊ฐ€์ž…์ผ Nisan 2007
176 ํŒ”๋กœ์ž‰459 ํŒ”๋กœ์›Œ
David Siska ๐ŸŒ
David Siska ๐ŸŒ@dsiskaยท
@_NogoodBoyo @evacide @peggymel2001 This is like โ€œencryption backdoor for the good guysโ€. It canโ€™t work, it will be hacked and real identities leaked. The bot ๐Ÿค– problem is real though and needs solving, but your proposal is not a solution.
English
0
1
4
78
David Siska ๐ŸŒ ๋ฆฌํŠธ์œ—ํ•จ
Matt HB - matthodg.bsky.social
Matt HB - matthodg.bsky.social@matthewhodgยท
Love that we solved low cost, high capacity public transport 40 years ago and just decided never to do it again anywhere in this countryโ€ฆ
Matt HB - matthodg.bsky.social tweet mediaMatt HB - matthodg.bsky.social tweet mediaMatt HB - matthodg.bsky.social tweet media
English
107
675
11.3K
926.5K
Monte Carlo Seminar
Monte Carlo Seminar@OnlineMCSeminarยท
This coming Tuesday, our seminar welcomes @ArnaudDoucet1 , presenting on "Schrรถdinger Bridge Flow for Unpaired Data Translation". Hope to see many of you here! Subscribe to our mailinglist groups.google.com/g/internationaโ€ฆ and calendar @gmail.com" target="_blank" rel="nofollow noopener">calendar.google.com/calendar/u/0/eโ€ฆ for up-to-date schedule!
English
1
1
26
7.7K
David Siska ๐ŸŒ ๋ฆฌํŠธ์œ—ํ•จ
Gabriel
Gabriel@gabelblingยท
Amazing graphical representation of a neural net, never seen anything like it.
English
47
3.4K
19.9K
1.6M
David Siska ๐ŸŒ ๋ฆฌํŠธ์œ—ํ•จ
Anna Stokke
Anna Stokke@rastokkeยท
โ€œThe single biggest thing we can do to encourage kids to do maths at higher levels is teach it really well from primary,โ€ Ashman said. โ€œFrom the ground up, from the first year. If we do that, students will learn the maths, see their success and become motivated.โ€ smh.com.au/national/nsw/wโ€ฆ
Anna Stokke tweet media
English
8
30
142
9.8K
David Siska ๐ŸŒ
David Siska ๐ŸŒ@dsiskaยท
@Almost_Sure Iโ€™m pretty sure gps works in airplane mode; when walking in the hills I have phone in airplane mode to save the battery and yet maps +gps give me the correct location.
English
1
0
1
79
Almost Sure
Almost Sure@Almost_Sureยท
How does it know? It was in airplane mode ๐Ÿค”๐Ÿง
English
1
0
0
618
Almost Sure
Almost Sure@Almost_Sureยท
So Apple photos tags this as in Stoke Newington. Interesting
Almost Sure tweet media
English
1
0
4
1K
David Siska ๐ŸŒ ๋ฆฌํŠธ์œ—ํ•จ
Skyline Scotlandยฎ๏ธ
Skyline Scotlandยฎ๏ธ@Skylinescotlandยท
THE WAIT IS OVER!!โ€จโ€จSkyline Scotland returns in 2025 to our NEW Event Centre in Nevis Range, Fort William.โ€จโ€จImagine 2023 but... BIGGER, BETTER and with more ADVENTURE!โ€จโ€จSAVE THE DATE!โ€จโ€จ๐Ÿ“† Saturday 20th September 2025โ€จโ€จskylinescotland.comโ€จโ€จ๐Ÿ“ธ No Limits Photography
Skyline Scotlandยฎ๏ธ tweet mediaSkyline Scotlandยฎ๏ธ tweet mediaSkyline Scotlandยฎ๏ธ tweet mediaSkyline Scotlandยฎ๏ธ tweet media
English
0
3
10
957
David Siska ๐ŸŒ
David Siska ๐ŸŒ@dsiskaยท
If you connect LLMs to โ€œrealityโ€ more effectively during training theyโ€™ll produce better answers. For code and maths you can easily automate this. For other domains though?
Jonas Gehring@jnsgehring

LLMs for code should do much better if they can iterate on tests -- but they don't. Our new work (RLEF) addresses this with execution feedback at RL *training time* to use execution feedback at *inference time*. arxiv.org/abs/2410.02089 is just out! 1/6

English
0
0
1
89
Jordan Taylor
Jordan Taylor@Jordan_W_Taylorยท
Forget electricity: Think process heat! Here's the market for industrial process heat across Europe. Taken together it's bigger than all continental electrical generation. If we want to decarbonize, and not de-industrialize, we need to supply all of this without burning stuff.
Jordan Taylor tweet media
English
42
90
447
36.9K