Ray DelVecchio

@simonw I like using it for defined roles/tasks & spending 10-20 manual iterations to dial in the prompt/context combo. Clear instructions are the hard part, then it's like having the best employee. AI will be a godsend to creative introverts who don't have the energy to manage people!

I'm beginning to suspect that a key skill in working effectively with coding agents is developing an intuition for when you don't need to closely review every line of code they produce. This feels deeply uncomfortable!

@petecodes That’s mostly who I work with. Def get a mix of the best clients and those who contact you every time things are slow. they get nervous and want an immediate solution to ranking first... tend to play dumb when you explain, and become experts if something is wrong lol.

I'm enjoying doing SEO work for a builder client There are so many keyword opportunities in unfashionable industries like the trades, it's unreal And their websites all need spruced up Stop just making stuff for your indie hacker mates 😀

@petecodes @John__McT I second this. Bought one like 10 years ago with pack of 100 blades and it lasted me 5 years before I had to buy anything else. I stopped using shaving cream too… just a dab of Castile soap does the trick IMO.

@John__McT Thanks man - looks ideal

Guys - what is a good single use razor? 🪒 I want something non plastic for avoiding the dreaded neck beard

@litcapital I feel like this is a great microcosm for AI in general. People laugh that it’s still wonky, and write it off l then a few weeks later people figure out it’s actually useful. That said, I am not a VR wearables guy since it’s seemingly a distraction from reality in front of you

Something raw about doing a live demo and not having it go flawlessly. Have to respect the move than choosing to run some pre-recorded demo Also probably drawing a lot more attention than they would’ve gotten had it been perfect

@ID_AA_Carmack Same x.com/raydelvec/stat…

I like em dashes, and I hate that some people now take them as evidence of text being AI generated.

@herrmanndigital @HiltonHotels This is the real crisis in America. Companies get so big they can literally not give a shit about boots on the ground issues. Its unacceptable. Just toss them into the list like Meta, Google, internet companies, etc.

Just spent my afternoon begging @HiltonHotels to show some grace after I accidentally booked the same hotel twice for the same nights. Their response? “That’s policy.” So now I’m paying for two rooms and staying in one. Love that for me. Seriously though, where did good customer service go?

It cost $0.67 to create it using OpenAI (o4-mini) to do the planning (integrated with existing chat API wrapper), and Claude Sonnet to write the code. Just amazing!

Just built a custom UI to use the new ChatGPT 'gpt-image-1' model via API in < 2 hours. Simple PHP/jQuery + cURL, running locally on XAMPP. Cost of image: $0.26.

A Google Business profile was wrongly suspended, and someone there (or perhaps AI support review bot?) finally listened and didn't make the appeal process a complete nightmare?! Perhaps my first unexpectedly positive experience with a Google product in 5+ years.

@florinpop17 - Desktop app + API - File Combiner for Context (BAT script > XML format) - Create plan (o3-mini, optional) - Ask Question (claude) - Update Code Manually - Test/Repeat I prefer this bare bones w/ Notepad++ as editor, and refresh context often so efficient (max is like $2/day).

What is your AI coding workflow?

New use of AI: Having a conversation with your dog about their misbehavior 🐶

@RDValerie they send “call me” texts which is the worst with no reference (99% isn’t urgent). So I push the limits of not responding for 1-2 days as a passive pushback, if I assess it’s not a pressing issue. I do DND from 830pm to 1130am too. Ultimately, would love to throw phone in ocean.

@RDValerie Yes 100%. I’m an introvert and tend to hyper focus, so even thinking that I need to respond when I see a text takes up mental space and hard to just let that go. I do tell my clients to not expect a response within 24 hours since I batch tasks…

With increased communication via technology, I feel like a lot of people, I find most to be honest, don't adhere to proper communication etiquette. One of the things that I miss the most from when phones and communication was simpler, is not always being available to people (and I know I don't have to respond), but I don't want to even receive communication at all times of the day. I value peace and quiet at certain times, and I shouldn't have to silence my phone for that as I need to be available to other people. but even if I do silence my phone, I will still see the notification. I have clients who text me evenings and weekends, and I had a sales rep that is my go to for skincare text me at 10:45 this past Friday night. While many people normalize texting, and some text with their clients if they are coaching and nothing wrong with that, I feel it should be reserved to friends/fam/acquaintances. I prefer emails and phone calls with respect to professional matters. I will have to set stronger boundaries around this. It is really important for my mental health and piece of mind to have downtime. I already work from home, so I have less of a boundary with work as I work evenings and weekends, but I do so if I choose to, and will work on reports so I'm not communicating with clients directly. I wish most people had a sense of appropriate communication etiquette, but I find that many do not. Does anyone feel the same?

OpenAI drops a phenomenal new image generation tool, and 95% of the posts are anime. Guys, anime sucks. The only adult cartoon worth watching is South Park. But I follow all the anime nerds because they are AI geniuses, as per @levelsio recommendation. 😆

@adgmodular @bentossell You have nailed their market. I got into web design with WordPress, and it has many flaws, hasn’t adapted well to more modern tech. But it ultimately lets you 100% customize. And in theory AI can walk designers through the WP tech setup steps and write custom function/CSS.

I love it, I came into web design late and had to witness the majority of my industry pretending Wordpress was good, so Webflow felt like a revelation. The company has definitely become bloated and has all the wrong priorities though. It’s got limitations for sure, but for branding focused designers with no great desire to become developers, there’s nothing better.

omg webflow is so so shit seriously does anyone enjoy building with it? any decent sized projects using it with no trouble?

@DanCalle @karpathy I don’t like long chats since it does tend to mix up/hallucinate, specifically if I’m iterating on code/writing within a chat. I’ll ask LLM when I don’t like answer to figure out why it failed and how to instruct better. Then add that to my role system prompts for future.

@DanCalle @karpathy My quick solution to having up to date file context was create a BAT file combiner, where I paste directory path and it concats all files into 1 XML file which I copy/paste into fresh conversations. I refresh context often, and export long chats as MD for potential future use.

When working with LLMs I am used to starting "New Conversation" for each request. But there is also the polar opposite approach of keeping one giant conversation going forever. The standard approach can still choose to use a Memory tool to write things down in between conversations (e.g. ChatGPT does so), so the "One Thread" approach can be seen as the extreme special case of using memory always and for everything. The other day I've come across someone saying that their conversation with Grok (which was free to them at the time) has now grown way too long for them to switch to ChatGPT. i.e. it functions like a moat hah. LLMs are rapidly growing in the allowed maximum context length *in principle*, and it's clear that this might allow the LLM to have a lot more context and knowledge of you, but there are some caveats. Few of the major ones as an example: - Speed. A giant context window will cost more compute and will be slower. - Ability. Just because you can feed in all those tokens doesn't mean that they can also be manipulated effectively by the LLM's attention and its in-context-learning mechanism for problem solving (the simplest demonstration is the "needle in the haystack" eval). - Signal to noise. Too many tokens fighting for attention may *decrease* performance due to being too "distracting", diffusing attention too broadly and decreasing a signal to noise ratio in the features. - Data; i.e. train - test data mismatch. Most of the training data in the finetuning conversation is likely ~short. Indeed, a large fraction of it in academic datasets is often single-turn (one single question -> answer). One giant conversation forces the LLM into a new data distribution it hasn't seen that much of during training. This is in large part because... - Data labeling. Keep in mind that LLMs still primarily and quite fundamentally rely on human supervision. A human labeler (or an engineer) can understand a short conversation and write optimal responses or rank them, or inspect whether an LLM judge is getting things right. But things grind to a halt with giant conversations. Who is supposed to write or inspect an alleged "optimal response" for a conversation of a few hundred thousand tokens? Certainly, it's not clear if an LLM should have a "New Conversation" button at all in the long run. It feels a bit like an internal implementation detail that is surfaced to the user for developer convenience and for the time being. And that the right solution is a very well-implemented memory feature, along the lines of active, agentic context management. Something I haven't really seen at all so far. Anyway curious to poll if people have tried One Thread and what the word is.

@amytheartist Yes and companies like Xfinity or Google are big enough to ignore your problems and know you’ll give up. On top of vague & purposely deceitful policies. Also product quality is shit. Grandma’s blender lasts 50 yrs yet a new one breaks in 5. Or steel razor vs plastic throwaways.