Daniel Saltzberg

3.4K posts

Daniel Saltzberg banner
Daniel Saltzberg

Daniel Saltzberg

@dargason

Data science, compchem, and structural biology. I turn numbers into other numbers to make digital arts and crafts.

Durham, NC Katılım Eylül 2009
1.7K Takip Edilen520 Takipçiler
Daniel Saltzberg
Daniel Saltzberg@dargason·
@iskander Do ASR with PAML, IQ-Tree, or whatever, to get an ensemble of plausible sequences. Run AF/Boltz on the output. Compare those structures to modern MHCs from your initial alignment. Note - I just made this up, have never tried anything like this. Sounds fun!
English
1
0
2
177
alex rubinsteyn
alex rubinsteyn@iskander·
What’s the best method (if there is one) for inferring the protein sequence at an unobserved / extinct node in a phylogenetic tree while conserving predicted structure?
English
7
2
14
3K
Daniel Saltzberg
Daniel Saltzberg@dargason·
@Michaelzsguo Cool! No novel trips currently planned, but you totally got my type. Definitely going to check it out!
English
1
0
1
36
Michael Guo
Michael Guo@Michaelzsguo·
If you love travel like I do, you know the pain: before every trip you're deep in Google, TripAdvisor, 小红书, 携程, reading strangers' tips, comparing restaurants, checking transit schedules. You easily spend days, sometimes weeks, on research. Travel is supposed to be leisure and relaxation, not a second job. So I built trip-optimizer, an open-source CLI that automates all that research, inspired by @karpathy 's autoresearch pattern. A few commands and you're done: ▎ trip-optimizer init "China 2026" ▎ trip-optimizer run ▎ trip-optimizer plan --pdf It autonomously researches your destinations, scores your itinerary across 7 dimensions (food, logistics, experience, budget...), and iteratively improves it — swapping in better restaurants, fixing timing, adding hidden gems locals actually recommend. Bad changes get reverted, good ones accumulate. Like gradient descent for trip planning. At the end you get a beautifully formatted PDF organized by day: attractions, hotels, restaurants, transit details, budget breakdown, safety notesm; all laid out and scored against your preferred vibes and tastes. ▎ 支持中文 — research prioritizes 小红书、大众点评、马蜂窝 when you choose Chinese. check it out and start using it today at: github.com/michaelpersona…
Michael Guo tweet mediaMichael Guo tweet mediaMichael Guo tweet mediaMichael Guo tweet media
English
2
0
1
143
Daniel Saltzberg
Daniel Saltzberg@dargason·
@SurvivingGrady My favorite seat at Fenway is GS33, row 3, seat 17 - where the LF side wall and Monster come together. Noone in front of you - great legroom. Can stand and lean up against the monster. Prime spot to heckle the opposing left fielder.
English
0
0
8
457
Red
Red@SurvivingGrady·
Being a tall guy at Fenway is like taking a couple hammers to the knees but these are some of the best legroom seats in the house
Red tweet media
English
23
4
251
24.3K
Corey Howe
Corey Howe@design_proteins·
New personal best ipSAE of 0.942 de novo binder towards RBX1 Using ProteinHunter
Corey Howe tweet media
English
5
9
66
4.4K
Jeremy Horpedahl 🥚📉
Pop quiz: What US MSA has the highest 10th percentile wage, adjusted for cost-of-living using the BEA RPP?
English
83
1
24
9.1K
Michael Guo
Michael Guo@Michaelzsguo·
Me too. In a team environment, every team member can share the skills too so the agents can get consistent results across the board. thank you for all your practical tips by the way. I'm working on a side project to put OpenClaw 🦞 on a Raspberry Pi and motor with wheels, turning it into an AI Rover robot that can roam around the house, one of the first things I made sure is to design the 5 MD files as you shared. x.com/Michaelzsguo/s…
English
1
0
1
192
Cathryn
Cathryn@cathrynlavery·
Openclaw tip 🦞 Don't put skills inside each agent folder. I keep mine in a ~/.agents GitHub repo — skills, hooks, and prompts all live there. Every agent symlinks in the specific skills it needs. My personal computer does too. Same skills shared across Claude Code, Codex, every agent, every machine. One skill updated = every agent and computer updated instantly.
English
36
25
467
29.6K
Daniel Saltzberg
Daniel Saltzberg@dargason·
@jumperz Yeah, that's what I found. Would be helpful to specify by channel, but haven't figured out a work-around yet. Appreciate you sharing your build!
English
0
0
1
53
JUMPERZ
JUMPERZ@jumperz·
@dargason yes and no, requireMention is supported at the guild level (applies to all channels in that server) but not per individual channel
English
1
0
1
868
JUMPERZ
JUMPERZ@jumperz·
openclaw dropped a good update and this one is super important for your swarm, but half the features don't work until you manually activate them. >ACP thread-bound agents i was already spinning up subagents in Discord threads to handle tasks. it worked, but i had to manage it myself by checking if they finished, cleaning up dead threads, etc. now openclaw handles all of that. the agent opens a thread, does the job, reports back...i only show up when an actual decision needs to be made. → do this: set spawnAcpSessions: true in your discord channel config >external secrets management my API keys, anthropic, elevenlabs, brave etc, were just sitting in a config file in plain text. anyone with access to that file had the keys. now they live in a separate vault and only get loaded when the system actually needs them. → do this: run openclaw secrets audit — it scans your config and tells you exactly which keys are still exposed and where >agent routing CLI wiring agents to channels used to take a lot of time. now it's just bind/unbind from the cli, so onboarding a new agent takes a minute or even less.. → do this: migrate with openclaw agents bind --channel --peer you have to do all of the above because openclaw ships opt-in features, not auto-changes... anything that touches security, permissions, or routing requires you to do it manually, and these are super important for your swarm. PS: if you're not on Discord, the ACP thread stuff doesn't apply, but the other two do.
JUMPERZ tweet media
English
57
45
975
93.7K
Daniel Saltzberg
Daniel Saltzberg@dargason·
@rw_eevee @yacineMTB Yeah, I'm in biotech, which is just slower. Accomplishing a significant amount of novel science in year 1 is difficult. Cycle times much longer - though getting better. Can see where in tech it's much different.
English
0
0
0
14
Robot Eevee
Robot Eevee@rw_eevee·
@dargason @yacineMTB I don’t understand this mentality. Two years at a job is pretty good in tech. I literally never had a hard time finding “a job” and every job has been more interesting and higher paid than the previous one
English
1
0
0
15
kache
kache@yacineMTB·
i need to hire people but i also don't want to because i realistically only have two years of runway
English
63
1
307
21.1K
Daniel Saltzberg
Daniel Saltzberg@dargason·
@yacineMTB Just be honest. Any startup has this risk. Personally, I love the 'gotta get shit to work or I need to find a new job' motivation - I hate looking for a job. But if you do choose to hire, make sure you are able to fire.
English
1
0
0
118
kache
kache@yacineMTB·
am i being retarded? i just don't want to give someone a job where i'm not sure it's going to be stable long term. these are people's lives
English
85
0
115
10.1K
Daniel Saltzberg
Daniel Saltzberg@dargason·
@zaimiri @realJoJo Yes, claude has timed me out. I switched to minimax for daily driver. Bought the $200 yearly plan for 300 calls over 5 hours (1/minute). Also, I use aggressive context compaction. I have gemini flash or grok 4.1 fast optimize context when it gets >150k tokens.
English
0
0
0
52
zaimiri ✏️
zaimiri ✏️@zaimiri·
Day 1 of Openclaw: Spent the entire day setting up and getting security right. Ate through $50 in API credits in a single day and kept getting rate limited. Managed to set up the OAuth now with Claude Max to get around that. Also today I: - Had my Openclaw generate its own pfp with Nanobanana - Gave it its own Notion account to keep track of what we're doing - Moved over to Discord to build the Subagent Architecture - Moving over old Claude Projects to new subagents
zaimiri ✏️@zaimiri

as someone obsessed with optimising and automating every detail of their life idk why this took me so long about to 10x productivity and efficiency

English
122
12
607
96.9K
Evan Lazar
Evan Lazar@ezlazar·
Coin toss: the Seahawks call tails, its heads. The #Patriots win the toss and defer to the second half.
English
6
15
140
11.6K
Chris Arnade 🐢🐱🚌
Chris Arnade 🐢🐱🚌@Chris_arnade·
@TrajansS @ReubenR80027912 exactly. When I returned to my normal place in Seoul, after a few months away, a few pounds heavier, the owner looked at me and said, "You have gotten puffy". LOL
English
2
0
18
1.3K
Chris Arnade 🐢🐱🚌
Chris Arnade 🐢🐱🚌@Chris_arnade·
One thing you notice in much of Asia (Korea, Japan, Taiwan, China, etc) is how often people eat out, rather than cook, so much so that their kitchens are usually very small, relative to the U.S. So there is an abundance of inexpensive street food/casual dining (carts, izakayas, takeaways, food courts, night markets, etc.). Part of that is cities (even smaller ones) are built for that (mixed-use zoning allows all sorts of shops everywhere), health codes are less onerous so the barrier to entry is lower, and everything is more dense, so you have tons of inexpensive options within several blocks. Without that in the U.S., your options are eat at home (time-consuming, knowledge barrier to clear), or drive to the strip mall X miles away. Or, DoorDash it.
English
152
419
6.8K
708.6K
Eric Webb
Eric Webb@webberweather·
Here’s a quick look at snow totals across NC for this weekend’s snowstorm I’m going to work on making a nicer looking, more detailed map later today How much ❄️ did you or some of your friends/family see? I want to ensure my map for this snow event is as accurate as possible
Eric Webb tweet media
English
62
16
195
20.9K
Daniel Saltzberg
Daniel Saltzberg@dargason·
@TimBuckleyWX Seen a few very tiny snow crystals dancinging in the breeze here in northern Durham County. Kids are getting worried! Do we have hope?
English
0
0
0
531
Tim Buckley
Tim Buckley@TimBuckleyWX·
Here’s where that tricky dry slot has set up so far today.
Tim Buckley tweet media
English
18
4
72
13.3K
Daniel Saltzberg
Daniel Saltzberg@dargason·
@8teAPi Makes sense from LLMs trained on shitposts and internet philosophers.
English
0
0
1
19