San Francisco
- Tweets 21
- Following 104
- Followers 293
- Likes 365
Ethan Kam retweeted
can the web help AI make better predictions from wearable data?
rabbitholed this week after reading @Meta’s WearableQA paper.
tested an LLM with and without @p0 search + extract and saw a meaningful lift in questions re: heart/metabolic health and inflammatory response with web data included!
6
5
15
918
muse has spoken
Parallel 🤝 Muse
Parallel Free Search can now supercharge your Muse agent. Just paste this:
Install Parallel’s MCP via Streamable HTTP at search.parallel.ai/mcp - no API key or OAuth needed. Verify it connects and exposes web_search and web_fetch and prefer for web search
2
36
12,742
Ethan Kam retweeted
Opus 5.5 pro tip:
Use Parallel web search. It’s 🆓
Opus 5.5 is the best model at using the context retrieved from web search
Just run:
claude mcp add --transport http parallel-search search.parallel.ai/mcp
Claude Opus 5.5 takes the #1 spot on Parallel's Search Capability Leaderboard, with a +4.3 gain over the second-highest score from Opus 5.
20
38
761
99,767
Ethan Kam retweeted
Today we're launching Data Connectors in Parallel.
Your agents can now work with specialized third-party data through Parallel's best-in-class agentic web research APIs.
Our first partners are @AlliumLabs, @useapolloio, @Baselayerhq, @CarbonArcAI, @crunchbase, Faraday AI, @harmonic_ai, @MiddeskHQ, @Similarweb , @particle_news, @pensa_systems, and @Polymarket.
We're also launching free, opt-in access to biomedical sources including PubMed, ClinicalTrials.gov, and ChEMBL.
9
22
100
51,209
Ethan Kam retweeted
Claude Opus 5.5 takes the #1 spot on Parallel's Search Capability Leaderboard, with a +4.3 gain over the second-highest score from Opus 5.
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.
4
5
88
109,817
Finally a model I can relate to. congrats anthropic on discovering overthinking from first principles
At its default effort setting, Opus 5.5 delivers frontier results for a fraction of the cost per task, often beating other models running at their highest settings.
It also generates output more than 30% faster than Opus 5.
1
14
1,268
For every matchup, we ran five searches using @p0’s advanced mode:
1. the matchup itself
2. news, injuries and storylines
3. matchup previews
4. team strength, offensive/defensive efficiency and roster changes
5. preseason previews and season expectations
We deduplicated the results and gave Jev the results as context
4
1
26
6,542
We used portfolio Kelly criterion to size bets against a $100 bankroll, maximizing expected log ending wealth.
Using Jev’s probabilities, we evaluated all win/loss combinations, assuming independent games and no ties. We priced contracts at the robinhood ask + a 2¢ fee allowance, with a $1 payout on a win and $0 otherwise.
We optimized over whole contracts with a $3 minimum per selected position.
1
9
5,108
I had Jev predict every nfl game this sunday and walked away with a 44% profit
It found 5 bets with an edge. 4 won
$67.14 risked → $97 payout
We gave it fresh context through @p0 search, compared its win estimates on robinhood, and used the kelly criterion to size bets
54
21
922
139,939
