building cool datasets at @mostlyrightmd
Prague, Czech Republic
Joined October 2017
- Tweets400
- Following276
- Followers453
- Likes1.1K
shipped on mostly right: dataset pages aren't one generic grid anymore. the agent looks at your data and picks a view, cards for places, product cards with real prices, plain table when nothing fits.
don't like it? just tell it in the chat
new in the mostly right chat: a bypass mode next to ask and auto. the agent approves its own builds and edits inside that one dataset and just finishes the job.
it still stops before anything public, destructive or touching other data
one fix that works for us: make the agent leave its work outside its head.
plan, sources and evidence written down as it goes.
you can't read its thinking but you can read what it checked, and that's usually what you wanted to weigh in on
the kimi k3.1 gap between benchmarks and vibes is the story of the year. open weights near the top of the leaderboard… but people pick models by cost per finished task and rate limits, not by an index score.. or? 🤔
Interconnects Podcast #20: We got together to debate hard questions and messy realities after a crazy, somewhat intimidating, week in the AI world:
- Why did Kimi K3.1 Pro have a mixed reaction with such incredible benchmarks? Will it be adopted?
- Were we wrong about the risks of open-weight models and cybersecurity?
- Why do cyber experts discuss recent events so differently than AI experts?
- What comes next in the few months? What will change our minds?
Chapters:
00:00 Kimi K3.1 Reaction
08:15 Kimi’s open-weight cyber principles
11:23 You can hack everything era
15:54 PHP2Shell, LiteLLM, and the new internet
24:00 Abliterated APIs, safeguards, and trusted access
31:41 Anthropic not hiring juniors employees
36:16 Florian’s Muse addiction
45:30 Wrap-up
More of these soon!
Security week every month is the right takeaway imho.
When agents run on your own boxes all day, the boring stuff matters most: least privilege, isolated machines, no secrets sitting in an agent's context.
Yesterday I spent all day typing this into Claude Code:
"Do a full security audit of this website and server"
On every single VPS (and site) I have
I already had automatic unattended upgrades with auto reboot for security updates in Ubuntu (reboot only takes 4 minutes so I think it's fine every few weeks)
Also isolated each VPS on Tailscale with special security policies (generated by Claude Code! so it can not access any other VPS (in case it gets hacked) and not even access my MacBook or iPhone unless I talk to it first
Also lots of user permission stuff isolating stuff to www-data to avoid root access for web services
Not the most fun stuff but essential
I think in addition to @yongfook's "marketing week" we now have to add one week per month of "security week" because with the models getting better our pre-emptive defense has to get better too
Luckily for us I think the attack and defense capabilities of AI models seems to somewhat match? So as long as we keep typing "do a security audit" regularly into our clanker we can keep it safe
I hope 🫨
spent yesterday deleting old pages instead of building new ones. mostly right is about datasets built by agents now, and the site finally says so.
weirdly satisfying to make the product smaller
more and more of real AI work is decisions, not text.
classify, route, pass/fail.
most pipelines never needed a chatbot in the middle, just a fast, calibrated yes/no... or?
AI's biggest breakthrough? My macbook's now a backpack furnace, warming my spine on these miserable fall commutes. Progress! 🚀
!many pixels were harmed again 🫡
@helloiamvu built ridiculously fast movie search on top of our movies dataset last week
i went down a rabbit hole making the datasets on Mostly Right feel just as good as Mojirama 👇🏻
!warning: it is super adictive movie search
mojirama.com
...and here’s what’s underneath:
1.26m movie + tv records: plots, cast, ratings
ready for ur “what i could build with that” moment
your turn;
mostlyright.md/datasets/movi…
Mostly Right has a new front door.
One place to start with a question, or a dataset.
A little tour ↓
The dataset cards got a rethink too.
You can now see how often the data updates. From live streams to daily or monthly refreshes, alongside likes, views and usage.
Also in this release: a proper mobile experience, light and dark themes, and smoother discovery.
Have a look and tell me what feels good, and what still gets in your way:
mostlyright.md
Should I do the same? 🫣
A dataset a day until @mostlyrightmd hits 1k users 😁
Today: Airbnb in Prague!
14% of host accounts advertise 62% of listings. The top ten advertise 938.
Built from 29 Inside Airbnb snapshots, with CLEF analyzing listing text.
Story + dataset in replies.
Bobbby retweeted
There is no model involved while you type, to keep it fast. Every keystroke runs a normal search in memory: a keyword index that fixes typos, plus an embedding index for anything that reads like a description.
Try it here: mojirama.com
2/6