harrisonjackson
harrisonjackson on Hacker News
The amount of video being produced has never been higher, as well.
Rise of influencers and content creators - less hollywood gatekeeping and more commoditizing the distribution of content across YT, Tiktok, instagram. Big names moving out of networks and into distribution channels they have more control over - making more content, more money, and with less resources.
X4 is great - has usb-c charger and with the cover feels like an easy to pocket and bring everywhere reader. Does not fit on the back of an iphone. I assume the magnet layout works with popular android phones though.
X3 is also great, actually fits on the back of the phone with the magsafe - magnet is a _bit_ too weak. It does fall off in my pocket frequently but I haven't lost it yet. Does not have usb-c - has weird little magnetic 4 pin charger.
I will gift my X4 to my brother once I've loaded some more content to it.
I recommend this for anyone that wants to save web novels for offline reading https://github.com/lncrawl/lightnovel-crawler - Calibre will sync the epubs to the x4 and x3.
Use the crosspoint OS https://github.com/crosspoint-reader/crosspoint-reader
All that said... I do still use my kindle as default. More content accessible more easily, better syncing, backlight
I found that doing more physical projects helped me. Large woodworking, home improvement, projects. Built-in bookshelves, a huge butcher block bar top (with 24+ hours of mindlessly sanding), rolling workbenches, and lots of cabinets. Learning and trying to master a new skill, using new design software, filling the garage with tools...
First, the number one problem is when the context is compacted. Any "plan" claude has in memory will not survive this for a complex task. Any extra instructions or guardrails you've provided just in memory will not survive this.
Most of these tips are to deal with that.
- plans go to text, Claude already does this somewhat well but you should explicitly place them in your current working directory in a big CURRENT_TASK.md
- larger plans with multiple phases are split up into multiple TASK_#.md or STEP_#.md with a TASK_SUMMARY.md and those live in CURRENT_TASK directory
- every single step that can be broken down more should be explicitly done so - if it can't accomplish the step in "1 context" before compacting it will likely fail
- every single step should have acceptance criteria
- all acceptance criteria should be testable by Claude Code
- You must heavily reenforce in claude.md and in the plans to perform the verification before moving on
- For most large tasks, manually accept edits
Most critical: all acceptance criteria MUST be something claude code can run a validation step on itself - some examples...
- unit tests, simple and easy and sometimes worthwhile
- typechecking, linting, building - other automated cicd stuff
- claude code can push and monitor gh action (I do this for docker builds that don't run nicely in my local -comfyui with nvidia/cuda stuff or example)
I will even do something as wild as:
> commit and push then monitor the gh action that builds the services image then check the logs as it deploys and connect with kubectl and look for <error being fixed> then repeat until it is actually fixed
That might spawn a task that runs for ~30 minutes while it monitors the github actions. Frees me up for another parallel claude code task.
It's very good with github, aws, pulumi, kubectl, and other cli tools. There are decent MCP integrations with sentry and other tools that can be useful for validation, too.
A lot of that will depend on your business and engineering org... don't give it the keys to the castle lightly.
I rented a lightning on Turo and it was amazing - planned on getting one as my next truck. I would drive a CT depending on price but they just draw too much attention.
> This is fake news. Location was not available on any gray check account at any point. Furthermore, the DHS has only shown IPs from the United States since account creation.
- head of product @x
Not to discount the impact of foreign powers over social media but maybe don't spread this misinformation.
I'd like to believe that most actual people want to protect kids.
It's easy to write off corporations and forget that they are founded by real people and employ real people... some with kids of their own or with nieces or nephews etc, and some of them probably do really care.
Not saying character.ai is driven by that but I imagine the times they've been in the news were genuinely hard times to be working there...
> "A bunch of people right outside your house!!!"
because it aggregates multiple single person walking by notifications that way...
Something a less technical parent can wire up in the family room from a trusted brand without having to do a ton of research for not only reputable brands but also vendors on amazon?
Educational mission aside this is a good alternative to a chromebook.
If I was you, then I'd keep hustling with upwork and whatever else you need to while building and launching something meaningful on your own in the space you want to work.
- slap a founder title up on your linkedin (set the start date to now)
- ship something not terrible
- continue iterating on it while becoming a better engineer, product manager, designer, etc all on your own
- learn to use AI coding tools really well
- clone and enhance the features of competitors
- talk about it a lot
- go to conferences for the business sector and for the tech stack
- network a ton
Then apply for a job if you still want to in 1-2 years. You'll have met a lot of people doing that and can hit them up or apply to competitors in the space using your startup as the perfect showcase.
When they ask why you are quitting just say you are super passionate about the space but couldn't raise money or going on a solo founder was terrible and you want to join a team.
And just don't apply for jobs at FANG ¯\_(ツ)_/¯ there are so many other companies out there that need people with the type of experience you'd have after 6 months to a year of the above. They are looking for folks with FANG background, tier 1 college/other pedigree, or specific experience building the same thing.
Also, extrapolating usage of vscode seems pretty fair since they said "most"
The LLM isn't performing the desired task.
It sounds possible to cancel out the comments where reversing the labels swaps the outcome because of bias. That will leave the more "extreme" HN comments that it consistently scored regardless of the label. But that may not solve for the intended task still.
Very easy to launch any docker image, script, stack etc.
You can templatize any repo for a 1-button deploy with a render.yaml and a link in the readme. More products offering "self-hosted" should do this.
It is challenging for Zendesk to enforce or fix DKIM, SPF, and DMARC issues across all client domains so better to just ignore it :grimace:
Directory SSO: These are systems like Google Workspace or Okta, which maintain a central directory of users and their access rights.
Non-directory SSO: These are services like "Sign in with Apple" (SIWA) or GitHub authentication, which don't maintain such a directory.
This is really cool. Are the costs to run this mainly storage or how much compute is actually tied up in it?
The time/cost to download models on a gpu cloud instance really add up when you are paying per second.
> The fastest image generation model tailored for local development and personal use
Versus flux pro or dev models
Here is pride and prejudice and up the thread you can see another web novel example:
https://twitter.com/HarrisonJackson/status/18109373574214537...
ElevenLabs has so many great voice models but is super expensive. I want to experiment with some oss voice models and even train my own but not sure on a great starting point with that. Play.ht has some good voices, too.
Seeing some of the results here with the openai tts I will probably switch at least the narrator to use one of these to save some money.
I am interested in any tooling or AI they used to help automate this.
The most useful lua code that I've written was very basic openresty nginx scripting around SSL certificates
I like the elephant frankenstein character - goofy and ugly/cute. Design, color scheme, copy, and animations are clean :chefkiss:
The value prop is well highlighted - at least for someone that's been out of php dev for a while. This seems like a nice way to bootstrap something small. Getting started code snippet makes it seem quick and easy.
Now the ads seem much worse whenever I have a non-premium experience on the service.
The standard deduction is a lot for some people.