Not yet. Atuin has been on my list of things to try for awhile now. Now it’s pretty close to the top
0
0
0
I'm seeing more and more evidence that the last few years of college-student cohorts are absofuckinglutely cooked (as in, they just didn't LEARN anything)
covid and now this, one-two punch
covid and now this, one-two punch
I am in Madison, WI and in a coffee shop I overheard:
"Yeah, claude was down so I couldn't do any homework. That sucked. But I guess its good because when I finish my undergraduate and go to Law school I'll have to start learning..."
What the fuck
10
3
184
9,419
Plot twist - reading and writing code becomes a bit less valuable because of AI, but then the scarcity of people that can do it makes it valuable again
0
0
2
203
I am in Madison, WI and in a coffee shop I overheard:
"Yeah, claude was down so I couldn't do any homework. That sucked. But I guess its good because when I finish my undergraduate and go to Law school I'll have to start learning..."
What the fuck
"Yeah, claude was down so I couldn't do any homework. That sucked. But I guess its good because when I finish my undergraduate and go to Law school I'll have to start learning..."
What the fuck
138
27
1,885
77,030
We're still so small but I'm loving the team we've got at Superlogical. They're so talented and kind and love their craft. It's a different era of software development and our expectations and practices reflect that. I want to share a bit about how we work.
First, everyone is trusted to push to main. We do machine-aided code review pre-push and human plus more machine-aided code review post-push. Limited to no PRs.
Culturally, we've all agreed that if we ever hire someone who we DON'T trust to push directly to main, then they probably aren't the right fit for the team (so far, not an issue).
Some common sense here: if someone is stepping outside their lane or comfort zone, we'll open PRs. For example, I'm pretty trash at macOS/iOS UI, so I'll open PRs for UI-affecting functionality and ask for eyeballs/help. But for business logic and internals, I push to main.
Second, everyone can cut a release. Our release process is fully automated and when you start a release (for any platform) it triggers a Slack notification and a 15 minute countdown.
The 15 minute warning is a final gate that allows people to review plus collaborate on a changelog. During that 15 minutes, anyone in the company can pause the release rollout. After 15 minutes, the train leaves the station. You can't override the 15 minutes (except to pause). It gives everyone a chance to step in.
Third, we avoid synchronization through fast remediation rather than through planning. The server and clients on various platforms ship features as fast as they can. We use post-hoc human review and AI-assistance to identify feature mismatches across products ("Windows has support for X and Mac doesn't." or "Mac and Windows differ in their behaviors for Y in this way.").
This avoids expensive "sync" meetings and coordination. We trust that producers of features are mindful of what they're producing and we trust that adjacent product lines will self-synchronize quickly. In practice, we see feature parity across platforms within 24 to 48 hours.
All of this works on mutual trust. It ONLY works with mutual trust.
Everyone at all times has to trust that they've got each others back:
1. Push to main only works if you fix your mess, or if you don't, that someone else will quickly but that you'd do the same (and generally in equal amounts).
2. Anyone can deploy only works if you trust everyone to cut at the right time, with the right changelog, and that people will respect the 15 minute timer.
3. Sync through fast remediation only works if you trust other teams will pick up and integrate your work quickly, which only works if you would too. It also only works if you are fairly producing mismatches (e.g. not constantly).
We've got a great team. I know we're still small and that these practices will change in many ways as we scale. But this is already a markedly different practice than what I would've done 15 years ago.
In general, we are building from day 1 on a foundation of mutual respect, excellence, and trust. And it's been awesome.
First, everyone is trusted to push to main. We do machine-aided code review pre-push and human plus more machine-aided code review post-push. Limited to no PRs.
Culturally, we've all agreed that if we ever hire someone who we DON'T trust to push directly to main, then they probably aren't the right fit for the team (so far, not an issue).
Some common sense here: if someone is stepping outside their lane or comfort zone, we'll open PRs. For example, I'm pretty trash at macOS/iOS UI, so I'll open PRs for UI-affecting functionality and ask for eyeballs/help. But for business logic and internals, I push to main.
Second, everyone can cut a release. Our release process is fully automated and when you start a release (for any platform) it triggers a Slack notification and a 15 minute countdown.
The 15 minute warning is a final gate that allows people to review plus collaborate on a changelog. During that 15 minutes, anyone in the company can pause the release rollout. After 15 minutes, the train leaves the station. You can't override the 15 minutes (except to pause). It gives everyone a chance to step in.
Third, we avoid synchronization through fast remediation rather than through planning. The server and clients on various platforms ship features as fast as they can. We use post-hoc human review and AI-assistance to identify feature mismatches across products ("Windows has support for X and Mac doesn't." or "Mac and Windows differ in their behaviors for Y in this way.").
This avoids expensive "sync" meetings and coordination. We trust that producers of features are mindful of what they're producing and we trust that adjacent product lines will self-synchronize quickly. In practice, we see feature parity across platforms within 24 to 48 hours.
All of this works on mutual trust. It ONLY works with mutual trust.
Everyone at all times has to trust that they've got each others back:
1. Push to main only works if you fix your mess, or if you don't, that someone else will quickly but that you'd do the same (and generally in equal amounts).
2. Anyone can deploy only works if you trust everyone to cut at the right time, with the right changelog, and that people will respect the 15 minute timer.
3. Sync through fast remediation only works if you trust other teams will pick up and integrate your work quickly, which only works if you would too. It also only works if you are fairly producing mismatches (e.g. not constantly).
We've got a great team. I know we're still small and that these practices will change in many ways as we scale. But this is already a markedly different practice than what I would've done 15 years ago.
In general, we are building from day 1 on a foundation of mutual respect, excellence, and trust. And it's been awesome.
169
178
4,368
261,778
I don't think most people realize how powerful (and easy) LLM fine-tuning is. Here's what it gets you and how to do it locally for free
4
1
9
714
So, in your example, you are taking a text entry that describes various events and when they happened and then mapping them to a JSON object, where the key represents when the event happened relative to date of the entry and the value is the description of the event.
That finetune is responsible for three different jobs at once: splitting events, labeling each by relative day, and emitting a particular JSON shape. It doesn't map neatly onto a single Jev request, because the event text is open-ended and the keys aren't a closed set.
But I wouldn't make the date the key. Instead, I'd invert it by making relative date a property of the event. Split the entry into events first (something a small LLM, fine-tuned or not, is good at), then ask Jev to classify each event into a fixed list, for example: Today (0), Yesterday (-1), A week ago (-7), A month ago (-30), or None of these. Weekdays and odd offsets, like 'On Sunday' in your example, go to the None bucket and escalate to an LLM. Code then builds the JSON from a defined schema.
That way the JSON is deterministic, the date label costs no output tokens, and any fine-tune is only for splitting events, which is a much easier ask of the LLM.
That finetune is responsible for three different jobs at once: splitting events, labeling each by relative day, and emitting a particular JSON shape. It doesn't map neatly onto a single Jev request, because the event text is open-ended and the keys aren't a closed set.
But I wouldn't make the date the key. Instead, I'd invert it by making relative date a property of the event. Split the entry into events first (something a small LLM, fine-tuned or not, is good at), then ask Jev to classify each event into a fixed list, for example: Today (0), Yesterday (-1), A week ago (-7), A month ago (-30), or None of these. Weekdays and odd offsets, like 'On Sunday' in your example, go to the None bucket and escalate to an LLM. Code then builds the JSON from a defined schema.
That way the JSON is deterministic, the date label costs no output tokens, and any fine-tune is only for splitting events, which is a much easier ask of the LLM.
1
0
0
26
Ok, so we're replacing the 1 LLM inference with two to three - 1 LLM, 1 Jev, and if Jev cant get it done, we do 1 more LLM inference. I'm honestly trying to think of scenarios where this would be preferable.
You'd have to make the "first" LLM substantially smaller than E2B. I suspect the simple task of splitting the events might be too much even for the base E2B to do consistently, let alone an even smaller model.
It also seems to be banking on being able to predict the most "common" relative dates so Jev's None responses are minimized.
Also, every inference is guaranteed to kick off a network request now.
You'd have to make the "first" LLM substantially smaller than E2B. I suspect the simple task of splitting the events might be too much even for the base E2B to do consistently, let alone an even smaller model.
It also seems to be banking on being able to predict the most "common" relative dates so Jev's None responses are minimized.
Also, every inference is guaranteed to kick off a network request now.
1
0
0
31
This is a good filter for people who I don’t want to work with. Please understand how computers work. This isn’t even about understanding Docker. Any software person should know how the layers work at a high level and this is a VERYYYYY high level.
A surprising amount of people don't know that Docker on macOS is actually a VM
190
176
5,769
500,929
@grok might be embarrassing myself here, but isn't the original post correct?
1
0
0
2,281
how is everyone using @superlogical's rex? like i work alone and don't use AI factories so no sub agents for me, so i'm curious what makes it compelling for solo devs like me
2
0
2
365
persistent sessions and tabs. across multiple machines if necessary. There's definitely a demographic who won't need it - those who find regular terminal tabs sufficient for juggling different tasks and who always work on the same computer
0
0
0
68
Right, but you can always back fill text fields by escalating to an LLM (a fine tuned one even)
It’s better than trying to fine tune an LLM to conform to a particular schema. That just feels wasteful
You could instead focus that effort on things that matter like response quality for a particular domain
It’s better than trying to fine tune an LLM to conform to a particular schema. That just feels wasteful
You could instead focus that effort on things that matter like response quality for a particular domain
1
0
0
17
Ok, let's walk through this. What "choices" would you offer Jev to achieve what we're doing here? Ie what would the request look like.
1
0
1
32
Wouldn’t it be easier to use a decision model like Jev if you want to output json with a particular schema?
1
0
1
27
Jev just makes a choice between two or more "options" that you give it. It cannot generate text or populate arbitrary JSON fields
1
0
1
34
Wow @superlogical Rex is very good. Thought I was in love with Herdr for life but I don't think I'm going back. Combining terminal and multiplexer in one app just simplifies things so much
0
0
7
451
More reason not to go anywhere near the Anthropic IPO
Andrej Karpathy:
“A 1B model can be much more effective than most current LLMs.
90% of current models waste a ton of capacity and money on remembering stuff that doesn’t matter.”
here is his 5 core thesis about Small LLM’s:
1. the intelligence was never the knowledge. Strip keep the reasoning. What’s left is the cognitive core, and it’s small.
2. the core might fit in ~1B params. Maybe less. “Plenty of room at the bottom.” We’ve been paying for a library when we needed a brain.
3. size stopped being the bottleneck. What makes an agent work is the tools it can call and how tight the feedback loop is, not parameter count.
4. intelligence lives in the system, not the weights. A raw LLM is a CPU with no OS. Small core + good harness > giant model alone.
5. you can own it. A model on your hardware is one no one can rate-limit, price-hike, or deprecate on you.
in 12 months the gap between teams shipping on-device and teams paying per-token frontier bills won’t be capability.
TwIL an 8.8B model at 90%+ of frontier isn’t a benchmark flex.
It’s the thesis Karpathy’s been repeating all year, made real.
study more on this below ↓
0
0
1
537
I just posted a video comparing 5 different Cloud Providers. To say that I've received some flak for it would be an understatement. The issue was exacerbated by a huge unexpected delay between production and publishing.
youtu.be/YfahrTa3Yb4?si…
youtu.be/YfahrTa3Yb4?si…
How can @CodeToTheMoons recommend railway after everything?
Is this a joke? Not only did they leak customer environment variables but they also had this whole incident with @joshmanders like five minutes ago
youtube.com/watch?v=YfahrT…
1
0
2
319
I appreciate that you came out immediately and explained the timing
I hope you will forgive me calling you out based on my initial impression. Knowing the timing details and that the video isn't sponsored changes things.
My apologies for jumping to conclusions, and thanks for being chill afterwards
I hope you will forgive me calling you out based on my initial impression. Knowing the timing details and that the video isn't sponsored changes things.
My apologies for jumping to conclusions, and thanks for being chill afterwards
1
0
0
22
No worries - to be clear the recent security breach is a valid criticism! It should absolutely be taken into account by anyone deciding whether to use the platform.
0
0
1
18
How can @CodeToTheMoons recommend railway after everything?
Is this a joke? Not only did they leak customer environment variables but they also had this whole incident with @joshmanders like five minutes ago
youtube.com/watch?v=YfahrT…
Is this a joke? Not only did they leak customer environment variables but they also had this whole incident with @joshmanders like five minutes ago
youtube.com/watch?v=YfahrT…
1
0
2
512
Spelling correction on my original response - nothing -> noting. Very appreciative of the empathy. OOC what platform would you recommend that you feel has a solid security narrative?
1
0
1
26
Does anyone have a single GOOD reason to move away from IaC due to agents? We've always been on IaC, plan to stay on IaC, and I can't understand why agents wouldn't make IaC even better than it was before
i see people saying
they're moving to iac because agents
they're not using iac because agents
they're switching to a monorepos because agents
they're moving off monorepos because agents
you're all confused man everyone is doing random shit
32
1
251
21,757