Rina Dewi Indahsari, S.Kom., M.Kom. - The AI Business Landscape - B18EB & Umum 2026-06-02 S2

Main ideas / lessons from the video

1) Live comparison of AI image generation (group activity)

• Participants were organized into multiple groups (earlier stated as 10 groups).
• Each group generated an image using different AI prompts/tools (some tool names were unclear due to subtitle errors).
• The host(s)/speaker evaluated the results visually and discussed differences such as:
  • Realism of faces (e.g., “real life picture” vs cartoon-like)
  • Normality over iterations (longer/more detailed prompts produced more natural results)
  • Specific visual cues, including facial features and skin texture (e.g., pores)

2) General conclusions about tool performance (as discussed by speakers)

GPT / ChatGPT

• Repeatedly described as producing more realistic results.
• Often considered better at capturing activity/context and producing more accurate profession-related likeness.
• After updates (specifically mentioned: “5.1”), hallucinations/errors were said to be reduced, and visual quality improved.
• Can become messy/mismatched if prompt formatting is wrong (e.g., accidentally including orientation text in certain modes like “meta”).

Gemini

• Often described as good, but sometimes produces a flatter or less realistic style compared to GPT.
• In some anecdotes, Gemini was credited for strong detail in product/background changes.
• Frequently characterized as more cartoon-like than GPT in direct comparisons.

Other tools mentioned (partial fragments)

• Mentions like Cloud / Grok / Quora / others appeared in fragments.
• One participant highlighted Grok as strong for latest news/sentiment, while noting it may be expensive (though sometimes there are limited/free options).
• NotebookLM was referenced for generating slide content, with limitations.

3) Prompting and workflow matters

• Same prompt + different tool ≠ same output
  • Even regenerating inside the same app can produce different results.
• Prompt specificity is critical
  • If you want a different theme, the prompt must be written more clearly and with more detail.
• If results are wrong: safer strategy
  • Start a new chat rather than only tweaking an old one.
• Editing beats full re-generation
  • If supported, users can zoom/click and edit only parts (e.g., change shirt color) instead of restarting from scratch.
  • The speaker emphasized you don’t need to be an expert prompt engineer—intermediate skill may be enough.

4) Data privacy and enterprise governance (major instructional section)

• Don’t upload confidential company data blindly
  • Even if the tool is used in a “personal” or “company” context, uploaded data may be reused/consumed by others depending on the provider.
• Avoid sending sensitive info directly
  • You can request analysis/reporting, but avoid exposing full internal datasets.

Enterprise AI governance “pillars” (8 mentioned)

1. Business suitability: choose AI that matches the business use case
2. Data context: understand where data is stored and how an automated agent accesses it
3. Integrate the right ecosystem (e.g., Google + Gemini)
4. Security & privacy: protect confidential data from leaks
5. Workflow integration: design how steps connect (e.g., cloud → analyze → Gemini/GPT → output)
6. Human review: validate outputs; don’t treat raw results as final
7. Legal issues: copyright/IP risks when generating and selling outputs
8. Total cost of ownership: include training + integration effort + implementation time (not just subscriptions)
9. Vendor maturity: verify vendors meet governance and industry standards

5) “Survival in 2026” framing: adaptability over intelligence

• The speaker argued success comes from fast adaptation to new AI tools—not from being the “smartest.”
• Tool selection is not one-and-done; you must try and test multiple times to find what fits.
• Recommended approach:
  • Identify the business problem
  • Evaluate context and data quality
  • Pick tools best suited to the problem/context

6) Tool discovery strategy and example platforms

• A “tool list” website was referenced (example: “Tolify AI”), with unclear counts (e.g., “29,000…”).
• Example platform categories referenced (from fragments):
  • Video generation models/platforms (e.g., “Sora 2.0”, “Kling 3.0”, “Gemini Omni”)
  • “Node-like” combined-model/video provider references (mentioned briefly)

7) Learning resources and cost/plan strategy

• Courses/platform references appeared (e.g., “Corsif.io”) implying structured learning for different AI tools.
• Pricing strategy suggested
  • Prefer monthly subscriptions over annual, since models/tools update frequently.
  • Don’t overpay for many tools—pick one best-fit “premium brain,” and use others via free tiers/support.

8) Q&A highlights

PPT generation tools

• NotebookLM
  • Can generate slides but has a 15-slide limit
  • PPT content may become non-editable images
• ChatGPT / GPT
  • Can generate PPT content that is more live/editable

Local terminal installation vs web access

• Even if models are installed internally, they may still be accessed via APIs/providers.
• Updates may lag behind the latest versions.
• Internet connectivity can still create data leakage risks.

Academic research

• For citations/scholarly references: use “GPT” + “Scholar AI” option (as described)
• For discovering research gaps: use Perplexity
• For data analysis: use cloud approaches; “AI scholar” was also mentioned as an alternative if you want to avoid switching tools.

Prompt reliability / “AI always agrees”

• Participants noted the need to counter AI suggestions:
  • Ask for critique
  • Press for questions first
  • Avoid overly positive/agreeable answers

Buying AI accounts on e-commerce

• Warning about cheap multi-user/reshared accounts:
  • Accounts may stop working after a day
  • Can be unreliable
• Recommendation: buy only from trusted sellers

---

Methodology / instruction-like content (detailed bullet points)

A) How to choose AI for a business problem (process described)

• Identify the business problem (what you want the AI to achieve).
• Determine the quality and context of the data you have.
• Match tools to the problem + context (select best-fit AI for that use case).

B) How to improve results when image generation is unsatisfactory

• If the prompt isn’t yielding the right theme/outputs:
  • Write the prompt more clearly and with more detail.
• If prior context might be contaminating results:
  • Start a new chat.
• When you get a mostly-good output:
  • Use edit features to adjust parts (e.g., shirt color) instead of regenerating.

C) Enterprise AI usage rules emphasized

• Avoid uploading confidential company data unnecessarily.
• Prefer asking AI to analyze and produce a report instead of sharing entire sensitive datasets.
• Require human review before acting on outputs.
• Consider legal/IP risk before generating and commercializing content.
• Calculate total cost of ownership:
  • Training employees
  • Integration time
  • Implementation effort
  • Subscription fees (not the only cost)
• Ensure vendor maturity:
  • Good governance and compliance
• Design workflow integration:
  • Example: cloud analyze → integrate into Gemini/GPT outputs

D) Cost strategy for subscriptions

• Prefer monthly plans over annual to benefit from continuous updates.
• Don’t pay for everything:
  • Choose one best tool for core work
  • Use others via free tiers or limited support when possible
• Use free/limited models strategically (example given: Grok free with limitations for news/sentiment tasks)

---

Speakers / sources featured (as identifiable from subtitles)

• Mrs. Rina Dewi Indahsari, S.Kom., M.Kom. (primary speaker/lecturer)
• Mas Hendi (host/moderator at times)
• Mr. Edi (presenter for cloud + GPT + data workflow example)
• Mr. Hendri (spelling variant; referenced as moderator/participant)
• Mr. Agung Situmang, Mr. Agung Supriyo (participants)
• Mrs. Agnes, Mr. Michael, Mr. Nata, Mrs. Tias, Mr. Sakti, Mr. Ridwan, Mr. Vinsen (participants)
• Mr. Yosef (asked about AI for PPT/presentation)
• Mr. Michel (asked about data privacy/upload implications)
• Mrs. Yuliana (asked about installing AI/cloud terminal and data leak risk)
• Mr. Julius (prompting/countering AI responses)
• Mr. Hendi / Mas Andi (mentioned returning/continuing moderator role near the end)
• NotebookLM (tool referenced)
• ChatGPT / GPT (including “OpenAI”)
• Gemini
• Grok
• Perplexity
• Google / Google Gemini / Google ecosystem
• Voice/video/editing-related tools appeared as fragments (e.g., Suno / ElevenLabs / Gamma / CapCut-like editing tools)
• Node.i.io (example platform referenced)
• ToList/Tolify AI (directory/tool list example)
• Corsif.io (learning platform example)
• Video/productivity tools/models referenced as fragments (subtitle accuracy may be imperfect): ElevenLab, Rapid, Loveable, Kling, Sora 2.0, Sedence 2.0