Thread: Artificial Intelligence programs for research and business

As Terry Pratchett once wrote, “real stupidity beats artificial intelligence every time.

it has its place, for sure. But I wonder why you’re asking here. Go and ask ChatGPT or Grok and see what they say. ??

Originally Posted by Tins

As Terry Pratchett once wrote, “real stupidity beats artificial intelligence every time.

it has its place, for sure. But I wonder why you’re asking here. Go and ask ChatGPT or Grok and see what they say. ??

This is what ChatGPT said (it better not start calling me Lionel! ):

Hi Lionel — really thoughtful question. Short answer: **yes**, there’s real substance behind tools like Manus, Genspark, and Skywork — but your instinct is also right: there’s *a lot* of hype, mixed with real emerging capability. Whether they’re “top 5 AI programs” depends a lot on use case. And for academic-research plus productisation (thesis → journal article, book, course, app, webpage), there are meaningful trade-offs.

Below is a breakdown of what these tools *are*, where they shine / fall short, what people are saying, and some suggestions — plus some caveats. (Happy to follow up with how to use them for academic work.)

---

## What Are These Tools (Manus, Genspark, Skywork)?

Here’s a quick explainer of each, based on publicly-available information plus user reports.

### **Manus AI**

**What it is:**

* Manus is an autonomous multi-agent AI system. It doesn’t just chat — it *executes* tasks. For example, it can browse web pages, run code, build websites, do research, fill out forms. ([manusagent.io][1])
* It runs asynchronously in the cloud: you give it a job, it continues working even if you disconnect. ([manusagent.io][1])
* According to Manus’s site, it’s very planning-oriented, with sub-agents for different tasks (research, code, execution). ([manusagent.io][1])
* It’s claimed to have very good performance on the **GAIA benchmark**, purportedly beating several strong models. ([manus.so][2])
* Pricing: there are “credits” for tasks. On some tiers, you pay monthly; other tiers more expensive. ([Manus AI][3])
* There is a *free tier*: according to its website, you get 300 credits/day for free and a one-time bonus. ([manus.so][2])

**What people are saying (pros & cons):**

* **Pros / hype**

* Very powerful for automating complex, multi-step tasks.
* Because it “thinks + acts,” it can be a big productivity multiplier if you set tasks up well.
* Could (in theory) help with research, content generation, web building, code.
* **Criticisms / caveats**

* Some users say it’s *expensive* relative to what they can get done. ([Reddit][4])
* Others report stability issues or that tasks “botch” or cost a lot of credits. >

> “I’m I the only one not impressed … It botches every coding project … uses an INSANELY high amount of credits” ([Reddit][5])
* According to a Business Insider review, in some tests Manus made pretty big mistakes (e.g., created a website using an *existing* domain’s content, rather than truly original content). ([Business Insider][6])
* There have been concerns about pricing and “task limit” vs credit model. ([Reddit][4])
* Because it's relatively new and very ambitious, it's not yet perfect — more like a very smart intern than a full, flawless autonomous agent, according to some. ([Business Insider][6])

**Verdict (for you):**
Good potential for converting your PhD work into other formats (journal article, course content, webpage) — especially because it can do research + writing + automation. But it will require *very careful prompting and oversight*. It’s not “set it and forget it” AI (at least, not yet) for something as important as academic content. Also, watch your credit usage closely so your cost doesn’t spiral.

---

### **Genspark AI**

**What it is:**

* Genspark started more as an AI-powered search engine. ([TechCrunch][7])
* It uses multiple LLMs (both in-house and third-party, e.g., OpenAI, Anthropic) to answer queries. ([Genspark][8])
* Instead of giving a list of links (like Google), Genspark outputs **Sparkpages**: dynamically generated pages that synthesize information from across the web. ([TechCrunch][7])
* These Sparkpages often have a built-in chat copilot so you can ask follow-up questions / clarify. ([Genspark][8])
* It has a “Super Agent” mode: the agent can do more than search — reportedly can take on tasks, not just research. ([Forbes][9])
* According to its docs and others, you can build custom agents (no-code) to automate tasks. ([Reddit][10])
* Genspark has a “Hub” feature: for each project, you can keep your files, prompts, and agents in one place. ([Reddit][11])
* Pricing: There are paid plans — e.g., Lindy (third-party) cites a “Plus” and “Pro” plan with monthly credits. ([Lindy][12])

**What people are saying (pros & cons):**

* **Pros:**

* Very useful for research: Sparkpages can save a ton of time when doing literature reviews / research overviews.
* More flexible than just a “search chatbot” — with multi-step, agent-based workflows.
* The no-code custom agent builder is quite powerful; you can spin up an agent for very specific tasks.
* Strong funding and momentum: Genspark is clearly serious and growing. ([Forbes][9])
* **Cons / caveats:**

* Some users report crashes: e.g., someone said the tool crashed while building a website, then reloaded with gibberish. ([Reddit][13])
* According to Reddit, some “Sparkpages” generated by Genspark may include a popup (“user generated content – contact genspark”) — potentially not ideal for polished public-facing content. ([Reddit][14])
* As with many AI tools, the output still needs strong human validation, especially for academic or business-critical tasks.
* Credit / cost model: if you use it heavily, the cost could ratchet up, depending on how they charge per “agent” task or research job.

**Verdict (for you):**
This tool is *particularly appealing* for the research-to-product pipeline you describe: turning thesis into journal articles, research summaries, course content, etc. The Sparkpage + copilot model might help you synthesize academic material quickly, and the “Super Agent” could help with automating content creation (e.g., turning chapters into slides, course modules). But again, you’ll need to carefully check and edit the outputs (especially for academic rigor / citations).

---

### **Skywork AI**

**What it is:**

* Skywork describes itself as an “AI-powered office suite” with “super agents” that can produce documents, slides, spreadsheets, webpages, podcasts, and more. ([PR Newswire][15])
* According to their launch, they have a “Deep Research” agent framework for research-style information retrieval. ([PR Newswire][15])
* Their agents also integrate “modular command plugins (MCPs)” for web search, document analysis, image generation, audio/video, stock data, etc. (per App Store description) ([App Store][16])
* On technical side, their research is quite real: there are recent arXiv papers from the Skywork team (e.g., *Skywork R1V* for multimodal reasoning) showing serious modeling work. ([arXiv][17])

**What people are saying / early impressions:**

* **Strengths:**

* Very much structured around “office-style” productivity tasks: research + writing + presentations + spreadsheets. That aligns well with academic workflow.
* Their Deep Research agent could be particularly good for doing literature reviews, building structured reports, or long-form documents.
* Because they’re building with modular plugins, there is flexibility: you could potentially integrate data sources or tools relevant to academia.
* According to the App Store, you can generate slide decks, visual data (charts), and podcast scripts. ([App Store][16])
* **Risks / limitations:**

* As with any relatively new agent-based system, there will be the “AI hallucination” risk / factual errors, depending on how the research agent sources information.
* Depending on how “modular command plugins” are implemented, you might hit limitations if you require very specific external academic databases (e.g., paywalled journals) — unless they support those.
* Cost structure may be a concern once you scale up complex projects; deep research is expensive computationally.
* Because it's “all in one,” the UI / learning curve could be more complex than using a simple chatbot or search.

**Verdict (for you):**
Among the three, **Skywork** probably has some of the most relevant architecture for your use case (academic → products), because its office-agent structure aligns with document/slides/course creation. If you value a unified workspace (research → deliverable), it’s compelling. But it’s newer and may require some experimentation and oversight.

---

## Is the Hype Justified?

Yes *to an extent*, but with big caveats.

**What is real hype:**

* The narrative around “agents that do everything for you while you sleep” is definitely out there, and some marketing leans into that heavily.
* Benchmark claims (e.g., “state-of-the-art” in GAIA benchmarks) are impressive, but real-world performance can diverge from benchmarks — especially for nuanced academic tasks.
* Some early users are overestimating how much these tools are “fully autonomous” vs how much they need coaching, prompt-engineering, and oversight.
* There is recruitment / investor hype, which inflates expectations; some of the “ultimate AGI” framing is more visionary than practical for now.

**What is genuinely solid:**

* These are not toy chatbots: multi-agent, planning, and execution are real capabilities now.
* For research-heavy work, tools like Genspark and Skywork represent a meaningful evolution: summarization + synthesis + structured output.
* For content generation (web, course, writing), these agents could save **lots** of time — especially in the early draft phases.
* Given their funding and technical publications (especially for Skywork), these are not fly-by-night tools: they have serious R&D behind them.

---

## Risks / Things to Watch Out For (Especially in Academia)

Because your domain is academic research and then productisation, here are specific risks and mitigation tips:

1. **Accuracy / Reliability**

* Use the AI-generated content only as *first drafts*. Always fact-check, especially with citations.
* For journal articles or thesis-based transformation, don’t rely on AI to generate final citations — cross-reference with actual academic databases.

2. **Data Privacy / Ethical Concerns**

* Make sure what you feed into the AI (your thesis, unpublished research) is okay to be processed by the tool’s servers — check their privacy policy.
* If you plan to generate content for publication, verify license / IP implications. (Who “owns” the text generated by the AI?)

3. **Cost Management**

* Monitor how many “credits” or compute you are using per task.
* Do a small pilot project first (e.g., convert one chapter of your thesis into a journal article) to estimate cost.
* Avoid very large open-ended tasks without a clear plan, because you might overspend.

4. **Learning Curve**

* You may need to spend time learning how to “prompt engineer” these multi-agent tools to get good outputs.
* It’s likely not “write one prompt, get a perfect journal paper.” Think in iterative cycles: prompt → review → refine.

5. **Tool Lock-in / Vendor Risk**

* As these are specialist/high-growth startups, there’s risk: features may change, pricing may change, or your credit model may change.
* Consider exporting outputs in standard formats (Word, PDF, HTML) so you’re not locked into their platform.

---

## My Recommendation (for You)

Given your background and goals (PhD → article → book → course → app / website):

* **Start small**: Pick one “transform” task to test. For example, ask Manus or Genspark to convert one thesis chapter into a well-structured draft journal article or executive summary.
* **Use a combination of tools**:

* Use **Genspark** for deep research and synthesizing literature.
* Use **Skywork** for creating slide decks / course materials / structured docs.
* Use **Manus** for automation (e.g., web scraping, building a minimal website, generating code for apps).
* **Iterate**: Don’t assume the first output is final. Use the AI to *get you 80% there*, then refine manually.
* **Budget explicitly**: Decide how much you’re willing to spend per month on AI credits, and track how much value (in time saved, content generated) you get back.

---

If you like, I can run a **2025 comparison of the “top 5 AI agents/tools” for academic → product workflows**, with pros/cons for each (including cost, maturity, stability). Do you want me to do that?

[1]: #1 Manus AI Agent - Fully Autonomous AI That Executes Tasks While You'''re Away "#1 Manus AI Agent - Fully Autonomous AI That Executes Tasks While You're Away"
[2]: Manus AI：General AI Agent "Manus AI：General AI Agent"
[3]: Manus AI | General AI Agent That Delivers Results "Manus AI | General AI Agent That Delivers Results"
[4]: https://www.reddit.com/r/ManusOffici...ce=chatgpt.com "Manus AI Status Quo"
[5]: https://www.reddit.com/r/ManusAiAgen...ce=chatgpt.com "Not impressed."
[6]: https://www.businessinsider.com/manu...ce=chatgpt.com "I tested Manus, China's 'fully autonomous' AI agent. It's promising — but not ready to go solo yet."
[7]: https://techcrunch.com/2024/06/18/ge...ce=chatgpt.com "Genspark is the latest attempt at an AI-powered search engine | TechCrunch"
[8]: https://www.genspark.ai/spark/what-i...ce=chatgpt.com "What Is Genspark Powered By? - Genspark"
[9]: https://www.forbes.com/sites/annaton...ce=chatgpt.com "AI Agent Startup Genspark In Talks To Raise Over $200 Million"
[10]: https://www.reddit.com//r/aicuriosit...ce=chatgpt.com "Genspark Launches Custom Super Agent for Easy No-Code AI Building"
[11]: https://www.reddit.com//r/genspark_a...ce=chatgpt.com "Introducing Genspark Hub: One Space Per Project, All Files in One Place"
[12]: https://www.lindy.ai/blog/genspark-a...ce=chatgpt.com "Genspark AI Features Guide for 2025: + Top Use Cases | Lindy"
[13]: https://www.reddit.com/r/genspark_ai...ce=chatgpt.com "Crashes, then refuses to restart"
[14]: https://www.reddit.com/r/genspark_ai...ce=chatgpt.com "What’s with the pop up"
[15]: https://www.prnewswire.com/news-rele...ce=chatgpt.com "Skywork Launches Skywork Super Agents Globally: The "AI-Powered Office Suite" Built on Deep Research"
[16]: https://apps.apple.com/au/app/skywor...ce=chatgpt.com "‎Skywork on the App Store"
[17]: https://arxiv.org/abs/2504.05599?utm_source=chatgpt.com "Skywork R1V: Pioneering Multimodal Reasoning with Chain-of-Thought"

Originally Posted by Tins

Go and ask ChatGPT or Grok and see what they say. ??

Total game changer in my job. Significantly improves the quality and significantly reduces the time demands for non face to face work.

Originally Posted by Slunnie

This is what ChatGPT said (it better not start calling me Lionel! ):

I hate what it's doing to youtube etc. I asked a question on goggle related to hormone blockers and sexual health and next thing I'm getting attractive "women ""doctors"" telling me all about it. Thing is, these were AI, and seemed to be giving what it thought I'd want, rather than what I needed. But, I was talking to someone doing medical research into the different effects of drugs on different illnesses. He had all the sources, but as you can probably imagine the workload was huge, and would have taken him months. AI did it in maybe two hours, and he said it was as accurate as he would have been.
Like I said, it has its place.

Hello All,

Using my words in creative ways - aye Slunnie! Interesting result though.

Tins, you wrote ... "But I wonder why you’re asking here. Go and ask ChatGPT or Grok and see what they say??" The reason I asked here on AULRO is that I wanted to tap into real users of AI. In my research I read some rather caustic reviews about some of the AI products. Some reviews included a more recommended product with the name highlighted in-text when no other words were highlighted. Then I glanced at the URL and found the "Totally Honest Review" was generated by the same company that recommended the better product. "Totally Biased Review" would have been a more accurate title. Also, online "influencers" get paid to promote certain products and their "Honesty" is somewhat tainted by their own pecuniary interests.

Some Findings of my preliminary research
I have been researching a couple of topics for over a fortnight. Therefore, I have done pretty extensive grunt work and have subsequently developed a level of understanding towards the topic. Also, an ability to approximate accurate from inaccurate information in relation to each topic has resulted from the grunt work. This provided me with some insight into how to word AI tasks. Some of the results were really amazing in the quality of their end products. Some were amazing in their inaccuracy and shoddiness.

I suppose it goes back to the old saying about computer data..."garbage in results in garbage out".

Kind regards
Lionel

Hello All,

I did trial one of those three previously mentioned AI programs. The trial was done after reading multiple online reviews and YouTube reviews. This is before the ChatGPT results were posted here on this thread.

I flexed the program's muscles on doing a multi-task process that I have been engaged with manually off an on for the past three or four weeks. Namely, going to a specific Australian Bureau of Statistic Local Government Area in Queensland. Doing a "Find in Page" function to locate the exact demographic cohort I was after. Then copying and pasting the ABS resulting number of the cohort and the percentage of the general poplulation with that LGA into an Excel spreadsheet. I repeated this process 78 times to cover each LGA in Queensland. I only got to the AI program to check the first five as a trail. The AI results showed that it visited the same ABS websites that I previously did in my manual research. I found that the AI program consistently generated incorrect data for each entry! For example, the AI result taken from the Brisbane LGA for my cohort of interest was down by more than three-thousand from the same ABS results! The AI reduction was 3,513 from the ABS webpage figure within the same URL sourced for the same cohort.

I always check data results and double check them. I suppose my years in high school transitioned from showing all mathematical calculations and learning to approximate before relying totally on electronic calculators. The lesson being, "do not automatically trust the data that the calculator generates. Is the calculator's answer close to your approximation or not?" Garbage in:Garbage out. My Year 12 class in New South Wales was the first where using electronic calculators was compulsory for the final tests. Not trusting electronic data and the exercising the ability to approximate still holds me in good stead all those years later. I am very glad that I did not unquestioningly include the AI results into a professional report. It would have been very damaging - reputation wise. Always check the facts!

Kind regards
Lionel