AGIcoming

I witnessed smth in Google which convinced me that WITCH companies can solve the problem of 'perennial problem of unemployable graduate'.

In Google Bangalore office, my team has few WITCH employees on contractual basis having salaries in range of 5-6 lakhs with work experience of 2-4 years.

While permanent Google employees are not allowed to use o1 models on official laptop, WITCH employees have no such constraints.

In last 3-4 days, WITCH employees are outperforming few permanent Google employees (having salaries in range of 60 lakhs to 1 crores) of my team.

This has convinced me that once Orion, Gemini 2 (embedded with Alpha zero, Alpha proof technique) etc. is released within year end, most dedicated WITCH employees (having salary of 5 lakh rs. ) can contribute in software development more than FAANG employees not using these AI tools.

2mo ago16K views

EnragedInterviewer

Wipro2mo

Wait until the bugs start coming in. The problem with AI right now is its garbage in garbage out. Hard to predict errors in ai generated coee

DecentWake

Student2mo

@EnragedInterviewer you sure about this? how is it garbage and if it's garbage then why does the code works? i guess you would be right if, your engineer is using ai code without understanding it. Isn't this same as using stack overflow (copy/pasting thing)

EnragedInterviewer

Wipro2mo

I meant if the ai is trained on garbage code , it will produce garbage code. Ive seen this in my project using copilot. So nowadays I only use it as line level autocomplete and not for copy pasting function implementations. In the next couple of years I foresee some companies losing lots of $ due to AI generated bugs

AlphaGrindset

Series A Startup2mo

Holy smokes For real? This is very big if your assessment is correct.

Isn't 3-4 days a small time to evaluate? Just curious because this is super exciting.

AGIcoming

Google2mo

Yes, I do agree 3-4 days is too short a time to judge something....but the trend I observed was too exciting to not tell people about it

Aloo

Microsoft2mo

You are not outside to write code, but paid to tell why to write that code or not write it at all. Unless you are in an org where LOC is what they measure.

Also why not use Google's models to do some work stuff?

Noctus

Stealth2mo

what's your prediction . New frontier models are gonna come after US election

AGIcoming

Google2mo

Sam Altman has recently hinted that Orion will release in winter which he said is going to be great. Google will release Gemini 2 this year which is power packed with reinforcement learning techniques. Anthropic next version and Grok 3 will also probably release by year end. This is going to start a new race.

And we must remember that not a single model has been trained on 1 billion dollar compute yet... Imagine how vastly capabilities are going to scale once models are trained on 1 and 10 billion dollars of compute.

My personal prediction is that by end of 2027 , AI will be able to do most "work from home " type jobs which does not require physical movement. And that will completely change job dynamics

Civ6Noob

Amazon2mo

What do u mean by outperforming?
Both get assigned different tasks right?, so we can’t compare . If they both get same task then what the hell is your manager doing lol