Sasha Goldshtein
@goldshtn
Followers
4K
Following
124
Statuses
8K
Software Engineer at Google Research. I work on Gemini factuality. Opinions my own. He/him.
Israel
Joined March 2010
Yeah. When I was first learning C#, I remember writing an interface called IAlgorithm which was basically a function pointer, and then another IAlgorithmProvider which had a single implementation and… you see where I’m going.
when you teach a kid the rules of chess, they'll start by making legal but basically random moves junior engineers are like this if they ever read a book about "design patterns" too early they'll just apply them randomly — singleton here, factory builder there, etc
0
0
2
RT @shubadubadub: How do we ensure humans can still effectively oversee increasingly powerful AI systems? In our blog, we argue that achiev…
0
18
0
@ZuckermanRoy לא פי 2. סדר גודל של 400 אלף דולר לדעתי. אני חושב שאם משקללים יוקר מחיה וכו׳ זה כנראה ריאלית 30% יותר? פה אתה יודע יותר ממני.
0
0
1
@ZuckermanRoy כן, אני מסכים שגם חצי מיליון דולר זה חיים טובים. אגב לקונטקסט, בגאמפא בארץ השכר הכולל הוא מאוד גבוה, ולא מייצג את מה שקורה ברוב ההייטק. לדוגמה, לבל 5 בישראל יכול לקבל טוטאל קומפ של מיליון שקל בקלות. ואפשר לחיות ממש יפה מארץ מהשכר הזה 😇
2
0
1
Today we published FACTS Grounding, a benchmark and leaderboard for evaluating the factuality of LLMs when grounding to the input context. The leaderboard is on Kaggle and we plan to maintain it and track progress.
1
8
26
RT @dipanjand: Throughout this year, we have had a razor focus on improving the factual accuracy of Gemini models' responses in various sce…
0
24
0
And also, our new Gemini 2.0 Flash model is top of the Vectara hallucination leaderboard. This is a 2.5x drop in hallucination rate since our last release.
0
0
22
Such a good model.
Breaking News from Chatbot Arena⚡ @GoogleDeepMind Gemini-2.0-Flash debuts at #3 Overall - a massive leap from Flash-002! Highlights (improvement from Flash-002): - Overall: #11 → #3 - Hard Prompts: #15 → #2 - Coding: #22 → #3 - Longer query: #8 → #1 - Overall style-controlled: #19 → #3 - Hard style-controlled: #25 → #2 The pace of improvement is absolutely astounding! Excited to see the new wave of applications powered by Flash. More analysis below👇
0
0
4
We have made more improvements since the previous experimental model 🔥 honestly, it is hard to keep track even internally. But this is a really good model on many many things.
Big news on Chatbot Arena 🔥 The new @GoogleDeepMind model gemini-exp-1206 is crushing it, and the race is heating up. Google is back in the #1 spot 🏆overall and tied with O1 for the top coding model! Highlights (improvement since gemini-exp-1121 in parentheses) - First place overall (2->1) - Tied with GPT-4o-1120 after style control (4->1) - Tied with O1 on coding leaderboard (3->1) - First place on hard prompts (2->1) Keep it up @GoogleDeepMind! The rate of progress is crazy. For analysis and to test the model, see below 👇
0
0
3