Join the discussion
Write your take first — we'll ask for email only when you're ready to publish.
- Hacker News
- > And it would go on like this for hours while you tried in vain to guess what the hell it wanted you to type, or you discovered the outdoors, whichever came first.
The part about Zork doesn't make sense to me. As I understand it text based adventure games are actually quite lenient with the input you can give, multiple options for the same action. Additionally certain keywords are "industry standard" in the same way that you walk using "wasd" in FPS games, so much that it became the title of the documentary "get lamp". Due to the players perceived knowledge of similar mechanics in other games you can even ague that providing these familiar commands is part of the game design.
It seems to me that the author never played a text based adventure game and is jut echoing whatever he heard. Projects like 1964's ELIZA prove that text based interfaces have been able to feel natural for a long time.
Text has a high information density, but natural language is a notoriously bad interface for certain things. Like giving commands, therefore we invented command lines with different syntax and programming languages for instructing computers what to do.
by trashb - Hi - Pranav from the ChatGPT Atlas team here. Appreciate the discussion and feedback here. Just wanted to correct a few specifics so there’s no confusion:
- We have tried to make Atlas a great way to browse the open web, and invested heavily in the search experience as we covered in the livestream. When I test typing “taylor swift" into Atlas I see links to the website in the autocomplete carousel, at the top of the results page, in inline citations within chat, and in the Search tab. We’re also working on making these faster and more prominent when we’re sure you want a link, for ex. you’ll see a large vertical list of links at the top of the chat response when you type in something like “gmail”. - Browser memories are opt-in, and by default we don’t train on the contents of pages you browse—even when memories are enabled. Keeping the Ask ChatGPT sidebar open has no bearing on what gets sent to us. Webpages are clearly displayed as attachments you can disable in the chat, and only sent to us when you submit the prompts. - Ask ChatGPT and Agent mode are not workarounds to user or publisher training settings. Even if a user has training enabled, we will still respect publisher preferences to block GPTBot in their robots.txt and do not train on that content.
by pranavvishnu - The Atlas implementation isn't great, but I'll pick something that tries to represent my interests every time. The modern commercial web is an adversarial network of attention theft and annoyance. Users need something working on their behalf to mine through the garbage to pull out the useful bits. An AI browser is the next logical step after uBlock.by xnx
- >We left command-line interfaces behind 40 years ago for a reason
No we didnt.
by protocolture - I opened ChatGPT on my Mac this morning and there was an update.
I updated ChatGPT and a little window popped up asking me to download Atlas. I declined as I already have it downloaded.
There was another window, similar to the update available window, in my chat asking me to download Atlas again...I went to hit the 'X' to close it and I somehow triggered it, it opened my browser to the Atlas page and triggered a download of Atlas.
This was not cool and has further shaken my already low confidence in OpenAI.
- People are misinterpreting the gui/cli thing.
There's 2 dimensions to it: determinism and discoverability.
In the Adventure example, the ux is fully deterministic but not discoverable. Unless you know what the correct incantation is, there is no inherent way to discover it besides trial and error. Most cli's are like that (and imho phones with 'gestures' are even worse). That does not make a cli inefficient, unproductive or bad. I use cli all the time as I'm sure Anhil does, it just makes them more difficult to approach for the novice.
But the second aspect of Atlas is its non determinism. There is no 'command' that predictivly always 'works'. You can engineer towards phrasings that are more often successfull, but you can't reach fidelity.
This leeway is not without merrit. In theory the system is thus free to 'improve' over time without the user needing to. That is something you might find desirable or not.
by PeterStuer - Being "anti-web" is the least of its problems.
This thing is an absolute security nightmare. The concept of opening up the full context of your authenticated sessions in your email, financial, healthcare or other web sites to ChatGPT is downright reckless. Aside from personal harm, the way they are pushing this is going to cause large scale data breaches at companies that harbour sensitive information. I've been the one pushing against hard blocking AI tools at my org so far but this may have turned me around for OpenAI at least.
by zmmmmm - This is amazing. My entire web browser session state for every private and personal website I sign onto every day will be used for training data. It's great! I love this. This is exactly the direction humans should be going in to not self-destruct. The future is looking bright, while the light in our brains dims to eventual darkness. Slowly. Tragically. And for what purpose exactly. So cool.by snorbleck