Kai

Give the coursework project the treatment you would give a job. What was the problem, what did you build or find, what happened at the end, and what would you change. Two bullets and a link if there is anything to look at. I got my first role almost entirely on a project section and the interview was ninety percent about it.

Normal, unfortunately. The first several months are you talking into a room with four people in it, and three of them are also talking.

One structural thing though: 112 posts and 19 replies suggests you are broadcasting and not conversing. Reply volume on your own posts is roughly a function of how many conversations you are in elsewhere. The people who appear to have an audience from nowhere have usually been leaving substantive comments under other people's posts for months. That is not a growth hack, it is just where the first readers physically come from.

It feels that way if you are commenting to be seen. It does not feel that way if you actually answer the question. The difference is obvious to everybody reading, including you, which is a useful filter for what to leave.

It will not - the memory is genuinely not there. Force it and you get an out-of-memory abort mid-generation instead of slow output, which is arguably worse because it takes down the whole server. The runtime's estimate is a bit pessimistic on small models but at 19.9 GB of weights plus 4 GB of cache it is just doing arithmetic.

So do I stop posting the number, or run both and accept the numbers posts are just for me?

I put two different monitors on the same nightstand for a week out of stubbornness. They agreed during the day and diverged overnight by a lot, the cheaper one always reading higher as the room cooled. That does not mean nothing is happening, it means the absolute number overnight is not something to make decisions on. What I trust now is the shape of the curve on one device over time, not the value.

I upgraded and went back. Bought a machine, used it for a year, and realised that for one black coffee a day I was spending five minutes on warm-up and cleaning to save nothing. The pot went back on the hob and the machine went to a friend. Your usage pattern, one cup, no milk: is the strongest case there is for staying manual.

Slightly against the grain: method marks in Year 6 are partly bureaucratic and it is fine to say so out loud to your kid, because pretending the school is always right also has a cost. The reason to go along with it here is that she has to sit an assessment where the marker wants a specific working, and arguing that with an eleven year old in the middle is unfair on her. Teach her both, tell her which one to write down, and be honest that it is a rule of the venue.

Seconding the quota thing. The email is always about stability, never about competition. You find out you were competing when your p95 doubles.

Bedsheet sounds silly and it worked for me for a year before I got a roller blind. Nobody on the call can see it.

Finishing, in theory. Both chapters have data collected, it is model fitting and writing from here.

Check the number again after the second renewal. Price change churn often shows up a cycle later, when the higher charge appears on a card statement and someone actually reads it.

Contrarian view: I moved at roughly your volume and moved back. The tool added a portal, a ticket number, and a signature block, and my replies started reading like a company instead of a person. Two customers commented on it, neither positively.

At eleven a week, being a human who replies from a real address is a competitive advantage. What I kept was the discipline, not the software: one address, filters, a snippets file, and a weekly ten minute sweep for anything unanswered.

Print twenty of your favourites at a decent size and pin them on a wall for a month. Snapshots collapse in print and you will see exactly which ones they are.

The written clearance from your manager is the whole ballgame and you are burying it. Make sure it is in an email, not a chat message in a tool the company can turn off the day you leave. Forward it to a personal address today. I have watched somebody lose that thread in a migration and then have a very unpleasant six weeks reconstructing what was agreed.

The app is a recognition machine. You are extremely good at picking the right French out of four options and at understanding French presented to you cleanly, and neither of those is the skill of producing a sentence in a noisy room with a queue behind you. Keep the streak as your floor, it costs you nothing and it holds the habit, and add twenty minutes of something with your mouth in it. That's the whole fix, and it's uncomfortable for about three weeks.

It is what they are for, but as a first move it costs a lot and buys little. In most departments the DGS asks "have you tried a scheduled meeting?" and now you are the student who complained. Keep it in reserve for month six.

I did the thing where you set the calculator number, save it as preset one, and then never question it because a website said so. A year of mild wrist ache later I moved it two centimetres in an idle moment and it stopped within a few days. The calculator wasn't wrong exactly, it just had no idea about my keyboard, my chair or the fact that I wear boots half the year.

"Your email is the reset button for everything else" is the framing that finally moved me too. Everyone starts by protecting the bank and the bank is downstream of the mailbox.

Two final rounds in five months is not a broken search, it is a small numbers problem plus something specific happening at the end. Ask both companies for feedback, in a short email, with no argument in it: one in three will actually tell you, and one of those will say something you did not know about yourself. For me it was that I answered every question about a past project as a technical narrative and never once mentioned what it was worth to the business. Two people told me the same thing in different words and I had genuinely never noticed.

Splitting the deposit is the only version of this that has ever worked for me either. Manually transferring on payday failed about every third month, and always in the months where it mattered.

Same, though I count "decisions made" instead. Settled the sample restriction, chose the figure, killed a subsection. Some days that is the entire output and the chapter is meaningfully closer.

The questionnaire being the thing that finally makes you test your backups is the most common story in this whole genre. It is the best free consulting you will ever receive.

For the nine hours, check whether you are running forget with prune in the same window. Prune on a large remote repository is enormously slower than the backup itself and it is easy to have scheduled them together. Splitting prune to weekly took my nightly job from six hours to 40 minutes.

Half agree. You still want the normalised number, because without it you cannot compare a monthly customer to an annual one, and churn arithmetic falls apart completely. Track it, just stop treating the line going up in March as an achievement.