NVidia GeForce RTX 50x0 cards - 70% performance increase, but AI > you

velk said:

Nah, the 'it's just a chatbot!!' is not even remotely accurate. In the coding 'game', it is now a fantastically valuable tool.

Is it replacing expert humans ? Not yet at least, but most people drastically underestimate the amount of expertise in any given field.

Is it replacing offshore developers ? I get an awful lot of A:B comparisons for exactly this case, and I'd say it's in the top 70% already, at a tiny fraction of the cost. If you are a junior dev doing contract work, it's time to start looking for a new skillset.

Click to expand...

It has replaced the need for entry level assistant positions at my company. It isn’t an experienced human but it’s more efficient than an entry level.

Khane said:

Wasnt suggesting it was a chatbot, just pointing out that it hasnt really changed, its just getting trained even faster and being given even larger data sets.

Click to expand...

Don't doubt this is true, just saying in my experience coding with AI assistants went from fancy auto complete to very thoughtful solutions real fast.

Latest Gemini builds can take plain english input and give regex output to parse log files or other raw data

So it's basically actual true AGI as far as I'm concerned now

Khane said:

Wasnt suggesting it was a chatbot, just pointing out that it hasnt really changed, its just getting trained even faster and being given even larger data sets.

Click to expand...

No, the approach for coding tools has changed quite a lot - they don't work the same way that the chatbots do anymore. Individual steps do, but there's a much larger shift in terms of orchestration and mixed-in non-llm tools, as well as changing ways that context inputs work to avoid the existing limits on model contexts.

For example, they used to be terrible at doing something like 'apply this change to all files in the project', because there were too many files with too much context for that to be possible for an LLM, so they'd just get part way through and then give up. What it will do now instead is break it down into steps like :
Check how many files there are
Do regex search for those files to find which ones need update
Create a temp file with the list of files that need to be worked on
For each of the files from the temp file, spawn off an LLM instance to actually do that change
Inspect the files for code/style errors
Run the unit tests, if there are any errors, work through them one at a time

That's even for basic changes - there's been an explosion in the adoption of MCP servers - every cloud provider and their dog is all-in on that. For some concrete examples on what that means - one of popular tools for designers is Figma - they'd make basically a pretty picture of what a mobile or web app is supposed to look like - with the MCP server for Figma integrated, tools like github copilot and claude can ask for the picture to be provided in a way they understand, and then produce an actual functional UI out of it that is identical to the design, and can be dynamically updated to meet new designs without user input.

These are all productivity enhancement tools for developers though - where the future is starting to look grim for junior devs is in the 'agent' space, where the paperwork and administration is bundled in with the actual coding. As a real life example, I recently ran across a problem ticket that was raised with a blurry screenshot of an error message and no other information. I raised an issue in the github repository for that application, pasted in the screenshot, and then assigned the issue to copilot. The result was, about 5 minutes later, a pull request from copilot with complete patch notes. If that ticket had gone to an actual human I'd give it a 95% chance that they just assigned it back to the call centre saying something like 'at least provide some details like when it happened or what the customer's login was'.

Think of the difference between, say, asking ChatGPT for a recipe for pancakes, and having it attached to some robot arms that can make you pancakes.

looks like the ai crash is imminent



micron shares are in freefall too, they essentially pivoted their crucial memory line away from consumers to sell to OpenAI and in the last few days OpenAI has basically said "oops, we can't afford anything anymore... help us raise some more money!"

lol, the scam didn't last very long at least. should see all the tech infrastructure companies freefall pretty soon, unless the jews convince Trump that somehow investing in OpenAI will defeat Iran or some retarded shit

The flaw was the memory companies not taking advance payment and only having letters of intent. You have to get paid before the debt collectors start circling.

spronk said:

looks like the ai crash is imminent



micron shares are in freefall too, they essentially pivoted their crucial memory line away from consumers to sell to OpenAI and in the last few days OpenAI has basically said "oops, we can't afford anything anymore... help us raise some more money!"

lol, the scam didn't last very long at least. should see all the tech infrastructure companies freefall pretty soon, unless the jews convince Trump that somehow investing in OpenAI will defeat Iran or some retarded shit

Click to expand...

Nah.

OpenAI was always going to fail. Very poor leadership, and their tech is 2 years old at this point with no windows for improvement. I predicted OpenAI would be the first to fail, and I predicted this a long time ago. There will be many other failures, but many successes. AI is still very much in its infancy, and YEAR over YEAR changes gives us new players and outlooks. I have never in my life seen tech evolve so quickly, and suspect this won't change much in the remainder of my life.

In other words, AI is still a giant gold rush, and any slack inventory that OpenAI wanted to buy will be purchased by someone else. RAM, NAND, all have top dollar customers to 2027. There are still more buyers than sellers.

Lanx said:

Click to expand...

I’m not worried that a machine will replace my job. I’m worried that a manager will THINK a machine can replace my job.

Until robotics advance enough.