Introducing deep research • Hacker News Board

Re: Introducing deep research

#441

Post by rvz » Tue, Feb 04, 2025, 9:28 PM UTC

blackeyeblitzar wrote: post #391

rvz wrote: post #10

It appears that OpenAI is in panic mode after the release of DeepSeek. Before they were confident in competing against Google on any AI model they release. Now they are scrambling against open-source after their disastrous operator demonstration and using this deep research demo as cover. Nothing that Google or Perplexity could not already do themselves. By the end of them month, this feature is going be added by a b…

I don’t think you’re comparing the right things here. This feature is more like Google’s Deep Research, which basically goes off and does a whole lot of search and compute to produce something more like a full research report. This has nothing to do with open weight models like DeepSeek (note: DeepSeek, Llama, etc are NOT open source). This feature doesn’t just require the research on the model but also enormous comp…

> This feature is more like Google’s Deep Research, which basically goes off and does a whole lot of search and compute to produce something more like a full research report.

Of course. It is in response to their disastrous operator demo which did not justify the $200 per month ChatGPT Pro subscription on top of the release of DeepSeek to make matters worse for them.

> This has nothing to do with open weight models like DeepSeek (note: DeepSeek, Llama, etc are NOT open source).

It obviously does. Even before they rushed this presentation, they made o3-mini available for ChatGPT free users so it in direct response to DeepSeek.

> This feature doesn’t just require the research on the model but also enormous compute. Plus anyone using such a feature for real work is not going to be using DeepSeek or whatever, but a product with trustworthy practices and guarantees.

Nothing that Perplexity + DeepSeek-R1 can already do.

So what is your point?

Re: Introducing deep research

#442

Post by infecto » Wed, Feb 05, 2025, 3:06 AM UTC

skywhopper wrote: post #339

infecto wrote: post #327

Earlier quoted context omitted.

We already have delivery drones though. I disagree though, it is useful as this problem has been whittled down and I think there is expectation that there will be continued effort. Its of course worth discussing but I find that for my workflows, I rarely encounter issues with hallucinations, they certainly exist but its gotten to a point that I don't have major issue with it.

At best, a proof of concept of experimental delivery drones exist, but only for small, lightweight items, and only in a few places, only in the right weather, and only if you place a target on your driveway and are there to receive the item in person, and all at the cost of a very high noise level. That's not exactly a real service.

You are sort of moving the goal post. The fact remains, drone delivery exists and is a solved problem. Major metro areas like Dallas, Texas have it through retailers like Walmart. Just because it does not meet your specific goal post does not mean it's a proof of concept.

Re: Introducing deep research

#443

Post by saagarjha » Wed, Feb 05, 2025, 8:10 AM UTC

justaj wrote: post #294

toomim wrote: post #246

Earlier quoted context omitted.

John Stewart had something to say about this: https://youtu.be/Byg8VZdKK88?si=pX1WbtRwZCBGpwHS&t=141

Without the tracking bits: https://youtu.be/Byg8VZdKK88#t=141

https://www.youtube.com/watch?v=Byg8VZdKK88&t=141s

Re: Introducing deep research

#444

Post by southernplaces7 » Wed, Feb 05, 2025, 7:40 PM UTC

HarHarVeryFunny wrote: post #385

gorgoiler wrote: post #293

For “deep research” I’m also reading “getting the answers right”. Most people I talk to are at the point now where getting completely incorrect answers 10% of the time — either obviously wrong from common sense, or because the answers are self contradictory — undermines a lot of trust in any kind of interaction. Other than double checking something you already know, language models aren’t large enough to actually kno…

> language models aren’t large enough to actually know everything I'd say they don't know anything . An LLM base model, before it is post-trained with RL, just has access to a sliced and diced corpus of human output. Take the contents of 4chan and WikiPedia, put in blender and mix and chop into "training sample" sized bites, then learn the statistical regularities of this blended mess. It is what it is - not exactly…

The amusing thing is that despite all you describe, and which many other people have described on this site and in others in great detail about the nature of LLMs, there are still many people who believe that these models essentially possess intelligence, and that it's little different from how a human mind expresses cognition.

I've seen this thinking to be especially prevalent among tech types (especially among many comments on this site), and more so than among average non-tech people I know.

It seems to be a reflection of a certain forced, almost ideological techno-reductionist thinking against the honestly complex and largely mysterious nature of consciousness. Many non-tech people on the other hand accept this mystery of consciousness and paradoxically are thus less likely to consider an LLM to be anything deeper than the clever but mindless pattern-matching trick that it is

Re: Introducing deep research

#445

Post by hmottestad » Thu, Feb 06, 2025, 1:17 PM UTC

hmottestad wrote: post #439

svara wrote: post #345

Earlier quoted context omitted.

> Gemini has had this for a month or two, Would have loved to try it when they released it, but I'm apparently in the wrong country. I think it's not available outside the US (?). OpenAI and DeepSeek have no such issues. It's a bummer really, I'm happy paying for this but they don't want me to.

OpenAI Deep Research isn’t available in Norway at least, or the rest of Europe basically :(

It seems to be available in Norway now.