Eric Deng
@ericcdeng
Followers
803
Following
6K
Media
92
Statuses
2K
helping agents go beep boop and click clack @google. Views are my own!
San Francisco, CA
Joined February 2014
RT @DynamicWebPaige: 🖌️🧵 Convert any painting (or photo!) into a paint-by-numbers template, or a counted cross-stitch pattern:. @GoogleAISt….
0
3
0
☁️☁️☁️☁️.
Introducing Chroma Cloud: an open-source serverless search database that is fast, cost-effective, scalable, and reliable.
0
0
3
Generate videos in just a few seconds. Try Grok Imagine, free for a limited time.
536
677
4K
Woo! Congrats @srikanthatcache and the Cache team!.
🚀 We’ve raised a $12.5M Series A at a $125M valuation. We’re solving a problem hiding in plain sight: overconcentration in company stock. We understand – selling means a massive tax bill. But, holding means massive risk. Cache has shown there’s a smarter way!. Cache Exchange
0
0
2
RT @saswat101: .@Spotify on @Waymo! You asked for more audio options, we listened. We're excited to partner with Spotify to offer another w….
0
9
0
I wrote up more details in this quick Substack here - There are more details on the issues and how to solve them as well as a full recording of this orbit with a hand breakdown for the real poker nerds.
2
0
3
Comet does still have gaps in its basic poker reasoning - the most basic example of this was when it thought that "3s full" would beat "Ks full" and a slightly more advanced knowledge gap was when it thought that the desired outcome of a "value bet" was to get others to fold.
1
0
0
Comet would sometimes misread cards completely - I wish I had KK instead of K7 here. Unfortunately Comet is also quite confident in it's read, INSISTING that the user had KKs in this hand even though the user's hand was visible this whole time.
1
0
0
Comet was confused by the UI in PokerNow, maybe because some of the actions (BET) were only available when it was your turn vs CALL, CHECK, and FOLD were always available. This led to some illogical reasoning and actuation issues like clicking CHECK to "enable the BET button"
1
0
0
Comet's poker reasoning was overall quite reasonable - it had basic knowledge of poker strategy and understood hand strengths, etc. In one instance it even demonstrated an ability to think about prior hands - this hand is "significantly stronger than previous hands".
1
0
0
Setup: I just ran Comet with a simple prompt on PokerNow's free-to-play "Friendly Sit and Go" for just one orbit (7 hands). I had to tweak the prompt a bit to get this to work (details in the Substack).
1
0
0
One Orbit of Poker w/ @PerplexityComet . Over the weekend I wondered how well Comet would be able to play poker out of the box. turns out it was really decent at basic strategy but it definitely ran into some challenges too. some details + examples in the🧵👇.
1
0
8
RT @jmj: 🚨 Dream job alert: We’re hiring an Investment Partner at Chapter One in San Francisco & there’s never been a better moment. SF is….
0
9
0
RT @AndrewYNg: The invention of modern writing instruments like the typewriter made writing easier, but they also led to the rise of writer….
0
279
0
RT @demishassabis: Official results are in - Gemini achieved gold-medal level in the International Mathematical Olympiad! 🏆 An advanced ver….
deepmind.google
Our advanced model officially achieved a gold-medal level performance on problems from the International Mathematical Olympiad (IMO), the world’s most prestigious competition for young...
0
760
0
RT @sh_reya: My hot take: PMs are the bottleneck in AI-first products b/c many engineers find "looking at the data" and doing evals "beneat….
0
35
0
RT @levie: The future of software is going to be about managing the work that AI Agents are doing. The UX will be about task management, re….
0
80
0
what part of the adoption curve is this?.
OH in San Francisco. Child: “What’s an Uber?”. Adult: “It’s like a Waymo, but with a driver”.
0
0
5
RT @BorisMPower: Yes - my advice to startups is prototype 10 product features - create an eval for 7 that don’t quite work yet and focus on….
0
31
0
RT @TheGregYang: a mathematical habit I find useful for product thinking:. in the limit, what does . look like?. for example, what will h….
0
141
0