I'm just a vibecoder trying to make my way in the universe. founder, USC RED, ex debt + structured finance at nmrk
Los Angeles
- Tweets 1,054
- Following 574
- Followers 157
- Likes 7,679
I was verbally assaulted at the gundo bonfire last night, someone asked if I was "non-mechnical". Need to brush up on die casting I guess 🤷♂️
1
64
I've built the literal Eye of Sauron for industrial real estate private equity, MCP server included obvi
1
53
MET-Rx bars from Chevron are like 15% of my diet, I am soooo cooked.
25/25 protein bars tested positive for heavy metals:
- Aloha Chocolate Chip Cookie Dough
- Aloha Peanut Butter Cup
- Kreatures of Habit Daily Bar Choc PB Banger
- Laird Superfood Double Chocolate Peanut Butter
- Clif Bar Chocolate Brownie
- GoMacro MacroBar Sweet Awakening
- Jacob's Protein Bar Chocolate
- Prima Cacao Ancestral Protein Bar
- Santa Cruz Paleo Chocolate
- RXBAR Peanut Butter Chocolate
- GoMacro MacroBar Smooth Sanctuary
- Lineage Protein Bar Chocolate
- Built Puff Brownie Batter
- Barebells Creamy Crisp
- Blueprint Macadamia White Cocoa Butter
- David Peanut Butter Chocolate Chunk
- Perfect Bar Dark Chocolate Chip Peanut Butter
- PROMIX Protein Puff Bar Blueberry
- RXBAR Chocolate Sea Salt
- KIND Breakfast Protein Dark Chocolate Cocoa
- Clif Bar Crunchy Peanut Butter
- Kirkland Chocolate Chunks Cookie Dough
- Kirkland Protein Bar Chocolate Brownie
- Rise Almond Honey
Lead: found in 23/25 (18 exceeded the Prop 65 limit, 1 exceeded FDA's limit)
Cadmium: found in all 25 (2 exceeded Prop 65)
Mercury: found in 2/25 (both exceeded the Prop 65)
Arsenic: found in 20/25
Full results on Oasis
Community note
The claim that all 25 bars "tested positive for heavy metals" refers to trace detections common in foods; per their data most were below FDA lead benchmarks and Prop 65 uses extremely low warning thresholds unrelated to acute safety.
oasishealth.app/science/25-pro…
fda.gov/food/environme…
oag.ca.gov/prop65/faqs-vi
85
What's fun about moka pots is they spit scalding hot espresso at you if you open the lid.
It's my second espresso of the day and second time I've burned myself. #imliterallyretarded
48
After a few weeks of having both Instinct and Poke, poke consistently lags instinct ~30 minutes for the exact same tasks.
70
What do LiDAR flyovers have to do with AI data labs?
Interesting discussion had on @TBPN regarding what the future holds for the booming data market.
Data research labs / marketplaces are making fortunes selling training data and reinforcement learning environments to the frontier labs. OpenAI / Anthropic frequently shelling billions to assemble datasets of ultra-hard coding tasks, human expert feedback, and niche vertical IP to (post) train their models.
There's debate over how these companies will endure 5 years from now, whether there's going to be demand for human curated data or a ceiling to scaling RL. Presumably AGI would produce it's own synthetic data for training or update models weights in real time.
My bull case is that labs will continue to pay vendors for refreshed / net new specialized data because intelligence, like Earth is not static and there will always be more nuanced ways to interpret environments.
It was coincidence that hours of compiling aerial datasets of counties in Ohio gave me the perfect pre-AI comparable:
businesses are still flying planes collecting aerial and LiDAR imagery.
On Google Earth / Maps, at the bottom of the page you'll see a thin label, like "imagery 2026, Airbus, Maxar Technologies", showing the vendor that produced the data rendered along with the satellite or plane / flight that captured it. Local governments, private companies, and researchers are constantly feeling the aerial data bottlenecks, even 1 day old is stale in the real world. For example, the images covering the Palisades Fire lag months, the last publicly available flyover dataset was produced in October '25, and apple is still showing pre-fire.
As technology generally improves (new satellites or data production methods), you unlock a higher definition of data and the recipes change, producing better AI models, maps, products, intelligence.
There is unlimited demand to continually scale intelligence and these Data and RL Co.'s are best positioned to provide data to improve nuance / definition for their respective domains.
you don't see Google flying their own airplanes.
2
242
Hard to deny global homogeny when there's an Auntie Anne's just minutes from the Chin Swee Caves Temple in Genting, Malaysia.
Travel 6,000 ft up the side of a mountain >1hr outside KL and you'll find outlet malls, casinos, and a starbucks inside a temple.
The neoliberal in me actually really enjoys the global city monoculture. I like the fact that I can buy a day pass at any nearby gym and later grab a poke slopbowl to hit my protein goals.
2
257
Today I used German MacBook, you'll notice lots of subtle key changes like Z and Y being switched.
I felt drunk while using it, couldn't type a single sentence without errors.
66
Agora v0, released in January. It's taken such different shapes for clients, almost ready to make v1, after these last few deployments.
Huge respect for platform engineers, wow is it hard to prune FDE work, have to be meticulous about what becomes part of the core product.
1
743
Favorite movie you like bc of your father
Opening weekend, 2010, laser show / 3D Chinese Theatre Hollywood. Peak.
1
1
234
Typically I only work in the terminal, but I just asked codex desktop to make a visualization so I can review changes I made to system prompts... I am shocked how good the output is, directly in chat and so clean
2
109
Excited to go deep into this harness, after building lots of agents with smolagents code-agent this seems like an interesting continuation of thin, python first tool execution.🤠🦋🇺🇸
73
A Claude max 5x subscription will only get you 15m 33s of Fable 5.
Unfortunately when session limit hit, multiple subagents failed in the workflow and I was also rate limited, leaving me with no usage or task recovery path.
When AGI is here, it will probably feel like this:
1
112
@spencerpratt LA Department of Mental Health Spent $4.7B...
Today the HHS DOGE team open sourced the largest Medicaid dataset in department history.
This dataset contains aggregated, provider-level claims data for a specific billing code over time.
For example, using this dataset, it would have been possible to easily detect the large-scale autism diagnosis fraud seen in Minnesota.
Download the data yourself:
opendata.hhs.gov
30
Since everyone is talking about clawdbot and Mac minis, here's my v1 Mini Farm from September 25. Believer in apple silicon way before the Claude army found its new toy.
I love the mini hype tho, the future is local
1
145
Face to Face deserves more hype.
Also grok imagine is improving (see old deadmau5 post), but could use better audio settings.
45
Kimi K2 thinking is scary good, might have so start brushing up on mandarin boys
Claude is still marginally better for the research agent I'm building, but it's impossible to justify the cost.
1
1
98
