GPT-6 Astra has created a lot of curiosity among people. Some see it as a major step forward in AI, while others are still exploring it. What is yo...
For further actions, you may consider blocking this person and/or reporting abuse
@shubhradev's your point makes a lot of sense let me start with it first.
The more capable a model becomes, the more plausible its wrong answers are. So capability and detectability move in opposite directions. A weak model fails in ways you spot straight away. A strong one fails in ways that survive review, because the output reads exactly like something correct. Every release makes the errors rarer and harder to catch at the same time, and reviewers pay definitely less attention to the latter.
On the jobs question, @merbayerp is right that he decides the architecture and takes responsibility. I'd add one thing from working in regulated environments.
Responsibility there isn't a personal virtue. It's a name on a document. NHS health IT has a named Clinical Safety Officer who signs off that a system is safe to deploy, and that signature belongs to a registered professional. A model can produce the assessment. It cannot be the accountable party, because accountability needs somebody who can be asked to answer for it afterwards.
So the honest version of the jobs question isn't whether AI can do the work. It's which parts of your job consist of being answerable for something. Those parts aren't going anywhere soon.
And the YouTubers are at least consistent. Selling shovels during a gold rush is the oldest reliable business there is. Anyone can research any AI tool, make a cool video and get a lot of views because we are in the AI era.😄
That’s a really good extension of the point.
The distinction between doing the work and being accountable for the work is probably going to become increasingly important.
AI may eventually produce an architecture, security assessment, migration plan, or even most of a production system that is technically better than what many humans would produce manually.
But when something fails at 3 AM, “the model generated it” isn’t an escalation path. 😄
Someone still has to define the acceptable risk, verify the evidence, approve the deployment, and ultimately put their name behind the decision.
And I think this also changes what makes an engineer valuable.
If AI keeps getting better at implementation, then blindly producing code becomes less valuable. Knowing what should be built, what must never be trusted without verification, where the failure boundaries are, and when to say “this is not ready for production” becomes much more valuable.
So maybe the interesting divide won’t be humans vs AI.
It’ll be people who merely consume AI output vs people who can actually validate and take responsibility for it.
And yes… meanwhile the YouTubers will continue selling shovels. 😂
You and @merbayerp go way deeper than my question!
😂 Fair point!
You asked “What do you think about Astra?” and somehow @ndcodes and I turned it into an engineering conference about accountability, production failures, and the future of work.
Occupational hazard, I guess. Give engineers a simple question and we’ll somehow find an architecture diagram hiding inside it. 🤣
But hey, you started it! 😄
Hahaha, guilty. 😄 And look closer, Divya. We didn't even answer your first question.
You asked what we think of Astra. @merbayerp and I both walked straight past it and went to accountability and production failures instead.
So, about your question. In my opinion, it's good, but I don't trust it yet. Not because it's often wrong, but because I haven't used it long enough to know how it fails. Oh yes! When it comes to AI and this new OpenAI tool, I'm still just an explorer.
Classic engineers. Given two questions, answer the one that wasn't asked and write four paragraphs about it.
Ask us something simple next time and see how long we last. 🤣
@ndcodes 😂 Exactly!
At this point, Divya should probably stop asking us questions and start giving us a word limit.
She asks, “What do you think about Astra?”
We reply:
“Before we answer that, we need to define accountability, establish the threat model, review the production architecture, identify the failure boundaries, and prepare a 47-page risk assessment.” 🤣
And after all that…
Divya: “Guys, I just asked if you liked it.” 😭😂
So yes, @technogamerz, we owe you an apology.
Not because we were wrong…
but because apparently neither of us is physically capable of giving a normal two-sentence answer. 🤣
Next time you ask us a simple question, please add:
“Mustafa and Nnamdi: maximum 50 words. No architecture diagrams.” 😂
I think Astra is really impressive, but for the amount of tokens it uses, I’m not sure its creativity has improved that much. 😹
It still doesn’t fully reproduce what I have in mind, so I usually need to refine and adjust the output myself.
It’s definitely powerful, though I’m curious to see how much this part improves next. 😸
I care less about what Astra can generate and more about how it fails. The failures that don't throw an error are the ones that can make it into production. That's the part worth watching closely as these models become more capable.
Exactly. The silent failures are the ones that worry me most. They can easily slip through testing and show up in production before anyone notices.
I think the interesting question isn’t whether AI will take people’s jobs.
It’s whether one person using AI effectively will be able to do the work that previously required five people. 😄
I’ve been using AI heavily in real software and infrastructure projects, and my experience is pretty simple: AI can write code, analyze systems, find problems, generate tests, and save an absurd amount of time.
But I still decide the architecture, define the constraints, question the output, test it, and take responsibility when something goes wrong.
So I don’t see Astra as “the developer replacement.”
I see it as another step toward a much more powerful developer.
And yes, there’s something wonderfully ironic about YouTubers making videos titled “AI WILL TAKE YOUR JOB!!!” while using AI thumbnails, AI scripts, AI research and AI editing to monetize the warning. 😂
Maybe AI won’t replace everyone.
But people who learn how to work with it are going to make life very uncomfortable for people who completely ignore it. 😏
Youtuber Will be youtuber!🙃
🤪
🤣
🙃
(◕ᴗ◕✿)
And do you remember GPT-4? Nostalgia for the days when AI used to put an emoji after every word!
I feel like AI is not mostly taking away people's job
it is gonna change the way of working.
yeah If people not evolved and remain working like a legacy systems legacy worker then for sure there job is in danger.
Do you know who makes the most money in the AI era? The people selling tutorials on how to use AI.🤣
Funny thing is, they turned a crisis into an opportunity—and me? Well, look at me! 😅
Hahaha, don’t feel too down. A gentleman loves wealth, but gains it the right way.(君子爱财取之有道)
सो धन धन्य प्रथम गति जाकी।
धन्य पुन्य रत मति सोइ पाकी॥
“那财富是有福报的，其最好的运用方式是用于布施（第一条途径）；那智慧是有福报的、成熟的，它致力于纯洁的功德与善行。那时刻是有福报的，当圣贤聚会（Satsang，与智者/善人相伴）发生时；那生命/诞生是有福报的，它对神圣与至高有着坚定不移的奉献。”
GPT-6 Astra, I have no idea - tell me all about it! ;-)
P.S. well, YouTubers will be YouTubers I guess - opportunistic by nature, and jumping at any chance to make a few $$$ ...
Haha, GPT-6 Astra is actually pretty useful.
It’s especially good at coding, reasoning through complex problems, and working with large amounts of context. I’ve also found it pretty good for agent-style workflows, where it can break a task down into smaller steps and work through them.
I’ve been using it for some of my projects lately, and the experience has been surprisingly good.😃
For me, AI is mostly an efficiency boost. I set up the workflow, then review the output myself.😏
Idiot me, still thinking they’re just tools for us 😂