Sean Oliver retweetledi
Sean Oliver
31.2K posts
Sean Oliver
@Sean_Oliver
Business Analytics at Microsoft📍Seattle. I write about Productivity, Personal Development, and Career. Note enthusiast. Ex Starbucks & Accenture
Read more here ➜ Katılım Ağustos 2006
1.1K Takip Edilen3.2K Takipçiler
this OpenClaw bot finds restaurants with ugly food photos, redrafts them as IG posts, and mails the owner a postcard...on autopilot.
here's how social-media agencies can use this system and land clients:
- scrapes every restaurant in a city in real time
- filters by review count + rating + last post date + photo quality
- pulls the strongest food photo from Google Maps reviews
- samples the brand palette from the restaurant's own visual identity
- AI-redrafts the photo into a 9:16 brand-matched Instagram post
- writes a postcard quoting a real reviewer + dish
- mails it to the owner by first name with a preview QR
every step from discovery to brand-matching to outreach is automated.
reply "GUIDE" + RT and I'll send you a free guide so you can build this too
English
an Anthropic safety researcher sat next to me at Blue Bottle and went pale
i was eating a sandwich and running my terminal on the laptop. she was on a zoom call. muted herself. took out one earbud.
"that's Claude running a live execution loop on a prediction market with real money"
yes. one repo. $25 a month. built it in a weekend.
she ended her zoom call.
"we have 40 people on the alignment team studying what happens when you let Claude make financial decisions on its own"
what did they find.
"it works. that's the problem. compliance won't let us publish the eval"
the repo: github.com/pmxt-dev/pmxt
1,500 stars. unified prediction market SDK. trades across Polymarket, Kalshi, and Limitless through one interface. the CCXT of prediction markets.
Claude read the SDK friday night. by saturday the full stack was running. kelly criterion sizing. whale divergence tracking. conviction scoring. spread capture on 5-minute BTC binary markets. 20 tools. 10 feeds. 24 wallets tracked.
"you're letting Claude find its own alpha from raw order flow. no rules. no hardcoded thresholds"
no rules. one prompt. Claude analyzes 147 whale events per session and decides which side of the market the crowd mispriced.
edge table live:
BTC+100K 5m +20c.
BTC+101K 5m +19c.
BTC<99K 5m +9c.
a green fill popped on the screen. +$140. she watched it happen.
this account runs the same logic. every trade on-chain. every P&L public:
@gabagool22?r=enables" target="_blank" rel="nofollow noopener">polymarket.com/@gabagool22?r=…
28,620 predictions. BTC 5-minute markets. pure spread capture.
another fill. +$91. she was still watching.
"how often does it trade"
55 times a day. out of 400+ windows. kills 85% before entry. only fires when conviction crosses threshold.
my setup:
> Claude - $20/mo
> VPS - $5/mo
> pmxt SDK - free
> Polymarket API - free
$239k volume. 82% win rate. sharpe 3.47.
+$15,171 from $900 seed.
i copied the setup here: @1743116" target="_blank" rel="nofollow noopener">kreo.app/@1743116
she stared at the equity curve.
"this is literally what our red team simulates. except you deployed it. with our model. on live markets. from a coffee shop"
Claude doesn't need an alignment review to place a trade.
she put her earbud back in. rejoined her zoom.
heard her say "we need to move up the autonomous trading eval. someone already shipped it"
Hanako@hanakoxbt
English
There should be a Tesla feature to play traffic and co-op and speed or smooth out the average traffic and stop and go based off of their relative position. If there were two Teslas back-to-back, they should be driving in such a way to average out traffic.
English
@videodb_io Wish it worked on windows. Won't let me select display
English
We just leveled up video infra for agents.
Give them video skills that normally take years to build.
Check out VideoDB Skills.
Naval@naval
It’s not about junior vs senior, it’s about “good with AI” vs “not good with AI.”
English
When presenting solutions, remember that complexity appreciation is for you; clarity is for them.
English
A core competency in a career is identifying (and offering solutions) to problems before your leadership does.
English
Silence about your work will be interpreted as irrelevance, not humility.
English
Once you clone yourself with AI everything changes
Most founders get stuck bc every decision goes through you
My Clone Yourself Checklist shows you which decisions to systematize 1st so your team runs without you
Comment CLONE and I'll share it. Follow me first or I can't DM you
English
Bill Pawliski from Prestige Fleet Services emergency-jumped my rental car at 4°F in Anchorage and refused payment.
I made my flight.
There are people who do the job and people who do MORE than the job. Bill's the second kind.
prestigefleetservices.com
English
Introducing autocontext: a recursive self-improving harness designed to help your agents (and future iterations of those agents) succeed on any task.
I built this for our clients with the intention of commercializing it but the community support around Karpathy's autoresearch convinced me to open source it instead. Our space is on the verge of something big and we want to do our part.
Andrej Karpathy@karpathy
Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual difference. I am mildly surprised that my very first naive attempt already worked this well on top of what I thought was already a fairly manually well-tuned project. This is a first for me because I am very used to doing the iterative optimization of neural network training manually. You come up with ideas, you implement them, you check if they work (better validation loss), you come up with new ideas based on that, you read some papers for inspiration, etc etc. This is the bread and butter of what I do daily for 2 decades. Seeing the agent do this entire workflow end-to-end and all by itself as it worked through approx. 700 changes autonomously is wild. It really looked at the sequence of results of experiments and used that to plan the next ones. It's not novel, ground-breaking "research" (yet), but all the adjustments are "real", I didn't find them manually previously, and they stack up and actually improved nanochat. Among the bigger things e.g.: - It noticed an oversight that my parameterless QKnorm didn't have a scaler multiplier attached, so my attention was too diffuse. The agent found multipliers to sharpen it, pointing to future work. - It found that the Value Embeddings really like regularization and I wasn't applying any (oops). - It found that my banded attention was too conservative (i forgot to tune it). - It found that AdamW betas were all messed up. - It tuned the weight decay schedule. - It tuned the network initialization. This is on top of all the tuning I've already done over a good amount of time. The exact commit is here, from this "round 1" of autoresearch. I am going to kick off "round 2", and in parallel I am looking at how multiple agents can collaborate to unlock parallelism. github.com/karpathy/nanoc… All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale of course - you don't just have a single train. py file to tune. But doing it is "just engineering" and it's going to work. You spin up a swarm of agents, you have them collaborate to tune smaller models, you promote the most promising ideas to increasingly larger scales, and humans (optionally) contribute on the edges. And more generally, *any* metric you care about that is reasonably efficient to evaluate (or that has more efficient proxy metrics such as training a smaller network) can be autoresearched by an agent swarm. It's worth thinking about whether your problem falls into this bucket too.
English
