I am curious what actually happens between two generations of AI models.For example, how do you go from Sonnet to Opus? Is Opus trained from scratch, built on Sonnet, or mostly the same model with more compute and training?And how do models like Astra suddenly make a big jump in some capabilities? What is stopping Anthropic, Mistral, or others from doing the same thing? Is the main difference...
Basically:
- /handoff file creates a short document with the important context from your current session and maybe next steps as checklist.
- You can then start a fresh session with /continue file
- You can also hand the work from Claude to ChatGPT, or the other way around. Very useful at time of session limits.
- Plus your handoff files becomes a useful piece of project memory that you can reference later.
I find this much more useful than /compact or /clear because the context is saved in something portable instead of being tied to one session and i've seen better results doing this every 20 messages or so than running long sessions.
For some reason, my brain follows it, and I fall asleep much faster. It works way better for me than box breathing or most other sleep tricks I've seen. Sharing in case someone else finds this useful.
> If your work isn't ready for users to try out, please don't do a Show HN. Once it's ready, come back and do it then. Don't post landing pages or fundraisers.
Session handling, auth, crypto, password hashing etc - all these are the exact areas where you should be the most allergic to rolling your own. Not because you're not smart enough, but because a simple bug like sanitizing in the wrong place and the failure is catastrophic like in this instance.
Use boring, proven, widely-audited solutions. Save your creativity for the actual problem you're solving.
For me, it is the shift toward thin, auto-hiding scroll bars. I see it on macOS, Linux (Mint), mobile phones and probably Windows too (though i haven't used windows in a while).Is this a cleaner look? I have always loved visible scroll bars because they act as useful guides for where I am on a page and how much content remains and just easy to drag. Now you have to hover over it first.I am...
It could be that I'm a bit old-school, but this really seemed to confirm that ready to play fun gameplay trumps realistic graphics any day!
Imagine just one link in a tweet, support ticket, or email: https://discord.com/_mintlify/static/evil/exploit.svg. If you click it, JavaScript runs on the discord.com origin.
Here's what could happen:
- Your Discord session cookies and token could be stolen, leading to a complete account takeover.
- read/write your developer applications & webhooks, allowing them to add or modify bots, reset secrets, and push malicious updates to millions.
- access any Discord API endpoint as you, meaning they could join or delete servers, DM friends, or even buy Nitro with your saved payment info.
- maybe even harvest OAuth tokens from sites that use "Login with Disord."
Given the potential damage, the $4,000 bounty feels like a slap in the face.
edit: just noticed how HN just turned this into a clickable link - this makes it even scarier!
Without updates, many sites will likely stop working with it soon.
Kiwi had some great features, like disabling AMP mode, rearranging the Chrome Store for mobile, and customizable tab layouts, etc. These features might interest others as well.
I did recently see this browser is unsafe when trying to open Gmail in it, so any chromium based update to date alternative there would be amazing!
The last time I used it one of the common hacks was adding 50 makers to a single app launch PH also openly condoned mass email blasts and tweets to drive votes which just rewarded whoever could push the hardest on promotion
In contrast Hacker News discourages asking people for upvotes and even treats it as a negative if you do That longterm focus on signal over hype is probably why HN still feels useful today while PH lost its way
- Big models like GPT-4 are split across many GPUs (sharding).
- Each GPU holds some layers in VRAM.
- To process a request, weights for a layer must be loaded from VRAM into the GPU's tiny on-chip cache before doing the math.
- Loading into cache is slow, the ops are fast though.
- Without batching: load layer > compute user1 > load again > compute user2.
- With batching: load layer once > compute for all users > send to gpu 2 etc
- This makes cost per user drop massively if you have enough simultaneous users.
- But bigger batches need more GPU memory for activations, so there's a max size.
This does makes sense to me but does this sound accurate to you?
Would love to know if I'm still missing something important.
Sam said yesterday that chatgpt handles ~700M weekly users. Meanwhile, I can't even run a single GPT-4-class model locally without insane VRAM or painfully slow speeds.Sure, they have huge GPU clusters, but there must be more going on - model optimizations, sharding, custom hardware, clever load balancing, etc.What engineering tricks make this possible at such massive scale while keeping...
My browser just froze after scrolling half-way. Not sure if this is something to do with the scroll fx but i really don't understand why this simple site is maxing out my CPUs.
When some users burn massive amounts of compute just to climb leaderboards or farm karma, it’s not hard to imagine why providers might respond with tighter limits—not because it's ideal, but because that kind of behavior makes platforms harder to sustain and less accessible for everyone else. Because on the other hand a lot of genuine customers are canceling because they get API overload message after paying $200.
I still think caps are frustrating and often too blunt, but posts like that make it easier to see where the pressure might be coming from.
[1] https://www.reddit.com/r/ClaudeAI/comments/1lqrbnc/you_deser...
Migrations randomly fail, schema changes are a nightmare, and your team forgets how SQL works.
ORMs promise to abstract the database but end up being just another layer you have to fight when things go wrong.
Theyve been hitting YouTubers like Mohak Mangal, Nitish Rajput, Dhruv Rathee with copyright strikes for using just a few seconds of news clips which you would think is fair use.
Then they privately message creators demanding $60000 to remove the strikes or else the channel gets deleted after the third strike.
It s not about protecting content anymore it's copyright extortion. Fair use doesn't matter. System like Youtube makes it easy to abuse and nearly impossible to fight.
It s turning into a business model: pay otherwise your channels with millions of subs get deleted
[1] https://the420.in/dhruv-rathee-mohak-mangal-nitish-rajput-an...
Heck they can make a back up directly to the cloud in addition to black box considering I'm able to watch YouTube in some flights nowadays.
If not there should be one as even my simple home wifi camera can record hours of hd video on the small sd card. And If there is, wouldn't that help to instantly identify such things?
Pretty sure this is against the Show Hn guidelines.
JsFiddle used to be my favorite for quickly testing out code snippets. It's a shame that due to Cloudflare hurdles, I've stopped using it and don't plan on going back.
It may not be much but as more websites and businesses lose genuine web traffic like this, Cloudflare might eventually listen and fix this mess.
