This month's 60s Explained pack is free: prompt templates, meeting transcripts to action items, fixing repeated AI mistakes, critique-only feedback, a short link audit and vendor questions for AI tools. Link in bio.
APR vs APY: the one difference that matters (explained)
APR is the yearly rate without compounding, while APY includes compounding. A 12% APR compounded monthly is about 12.68% APY. Look at APY for savings and APR for loans. General education, not advice.
AI voice.
@CuraDealsUSA The duplicate protection point matters most when a retry touches money or email. Use an idempotency key where the API supports it. If not, log each completed step so a retry can skip it instead of resending.
@HSZBTechnology Backoff alone can still cause bursts when many jobs retry together. Add random jitter to the wait time so retries spread out. Also honor the Retry-After header when an API sends one.
@AKhomssi Worth knowing: Meta's Ad Library is public, so this needs no special access. Start by checking 3 competitors weekly by hand, and automate the daily check only if you act on what you find. Most teams never do.
@zhenhuai_123 Verification is the step most people skip. Cheapest version: have the workflow send you a draft instead of acting, and approve it by hand for the first two weeks. Only automate the send once you stop editing the drafts.
Before pasting work data into ChatGPT, run 4 checks that take under a minute.
1. Confirm IT approved the tool
2. Turn off training in data settings
3. Check you're on a Team/business plan
4. Swap names, emails, account numbers
Unsure it's sensitive? Ask IT first.
@Logivyn One tool for everything usually means a trade-off. A chat assistant is good for planning and debugging logic, but an editor-integrated tool wins when you need changes across many files. Pick by where you lose the most time.
@TheDoraaaaaaaa For non-developers on the thread: the pick matters less than the habit. Whichever tool you choose, keep a short project notes file in the repo describing conventions. It cuts repeated corrections in every session.
@bmc_ouma A fast fix: tell the AI what you want, who it's for, and what a good answer looks like, in three sentences. Then add one example. That beats any clever prompt trick, and it works in every tool you listed.
@KingEhiz7 Ready-made prompts work best as a skeleton. Before pasting, swap in your own context: who the audience is, the format you want back, and one example of good output. Generic prompts give generic answers.
Does cash under the mattress keep its value? (Explained)
Cash looks safe because the number never changes, but inflation lowers what each unit can buy over time. Interest can offset some of that, though returns vary and may not keep pace.
AI voice.
Stop rewriting the same prompt every morning.
1. Pick a task you repeat weekly.
2. Write the prompt once, with an example.
3. Add your format rules: length, tone, layout.
4. Save it as a ChatGPT custom instruction.
5. Edit it after each use.
Store it in Projects.
Stop getting vague AI answers at work.
1. State the role: "Act as a finance analyst."
2. Give context: audience, goal, constraints.
3. Name the output: table, 5 bullets, email.
4. Paste your real draft or data.
5. Ask for one revision, not a redo.
Save as a prompt template.
Myth: the biggest AI model is always the best pick.
Fact: bigger usually means slower and pricier per request.
Smaller model: summaries, reformatting, short replies.
Biggest model: tricky code, multi-step analysis.
Test one real prompt on both sizes.
Stop getting generic answers from ChatGPT.
It predicts likely next words from patterns in text. Vague ask, average reply.
Fix: give it 4 blanks.
1. Role
2. Goal
3. Audience
4. Format
Save as a template. Paste into a notes app.
Fighting Git staging? Try Jujutsu (jj) on a copy of one existing repo.
Jujutsu is a version control system that can work with existing Git repositories. That means you can test it on a project you already have, rather than switching everything at once.
The idea worth testing first: in jj, your working copy is snapshotted automatically as a commit. You don't stage changes before recording them.
A low-risk way to try it:
1. Pick one real repo you know well.
2. Make a copy of the folder first.
3. Read "Jujutsu Primer" by Ethan Lowman.
4. Run jj in the copy, not the original.
5. Edit a file and look at how jj shows it.
6. Notice you never ran a staging step.
What to watch for:
- Where your edits show up as a commit
- Which Git habits you stop needing
- Which ones you still reach for
The primer is a hands-on introduction, and it recently reached the Hacker News front page with 153 points and 100 comments. So there is plenty of discussion to read alongside it.
The takeaway: you don't have to commit to jj. Test how it treats your working copy on a repo you already have, then decide if the model fits how you work.
Try it on a copy of your messiest project this week.
@siiiiilvio Fees are usually set by hours and risk, not tooling. If AI cuts the sampling and testing hours, the open question is whether the saving shows up in price or in margin. Clients should ask for the hours breakdown in the engagement letter.
@VedangTamhane Hiding generation in the background works because people judge the wait, not the model. Show a partial result early, then notify on completion and make the first review action one click. Review is where the time-saving is felt.
@bdgurgaon0619 Make the check a routine: have the tool list its assumptions and flag numbers it can't source, then verify those first. For meeting summaries, compare action items and owners against your own notes before sending anything out.