A few days ago, I analyzed First Mark's AI landscape:
AI Landscape Analysis
Daoming, WeChat Account: Daoming AI Lab
Is this the inaugural year for AI applications? A brief review of the 2024 landscape.
Actually, the 24 questions regarding 2024 raised by author Matt Turck are more worth discussing, many of which I have been contemplating recently. Regarding the author's discussion, due to IP considerations, I will not quote directly. Interested readers can visit:
Full Steam Ahead: The 2024 MAD (Machine Learning, AI & Data) Landscape https://mattturck.com/MAD2024/
Meanwhile, I selected four models to analyze "PART II" of the article—the discussion of the 24 questions. The models are Claude3 Opus, Kimi Chat, GPT-4, and Gemini 1.5 Pro. After several prompt iterations, GPT-4 only listed the 24 questions without summarizing the answers (OpenAI must speed up the launch of GPT-5), while the other three models listed all 24 questions and provided brief answers.
The model outputs are pasted below without any formatting adjustments:
1. Claude3 Opus (693 words)
- Structured vs. Unstructured Data: Structured data pipelines are cooling down, while unstructured data pipelines and Generative AI are heating up.
- Is the Modern Data Stack Dead?: The modern data stack is under immense pressure and may be replaced by the AI stack.
- Consolidation in Data Infrastructure: Big companies will get bigger and stronger.
- Competition between Databricks and Snowflake: Competition will intensify, and the launch of Microsoft Fabric poses a threat to both.
- Slow Transformation in BI: The Business Intelligence software industry is slow to transform, but Generative AI has the potential to bring massive change.
- AI Infrastructure Startups: Startups will reposition as "AI infrastructure companies," but moving from structured to unstructured data requires fundamental product changes.
- AI Hype Cycle: As hype continues, it's worth considering where we are in the cycle in 2024.
- 2023's Generative AI Boom: It might have been a "pseudo-boom," as much remains in the experimental stage.
- Differentiation in LLM Companies: Competitive differentiation exists among model companies; they won't be as commoditized as imagined.
- LLMs and SLMs: Large and small language models will coexist, and specialized models are rising; the future belongs to multi-model hybrid architectures.
- "Traditional AI" is not dead: It is complementary to Generative AI, and the future lies in hybrid applications.
- "Thin Wrapper" Startups: Startups need to differentiate through deep vertical integration.
- AI Agents and Edge AI: These are trends worth watching in 2024.
- Distance to AGI: Opinions vary on how far Generative AI is from AGI. Whether progress accelerates or plateaus remains to be seen.
- GPU Supply and NVIDIA's Valuation: An interesting topic. Long-term bullish on AI chip manufacturers.
- Open Source AI Community: The community has expanded too much and needs self-correction.
- Declining AI Costs: With the rise of open-source models, the cost of using AI is falling, which is a boon for businesses.
- Microsoft's Dominance: Microsoft has a comprehensive layout, but whether it has already won and if startups still have a chance is debatable.
- OpenAI's Strategy: OpenAI's strategic path and relationship with Microsoft deserve continuous monitoring.
- Enterprise AI Landing: 2024 is expected to be the year for enterprise-level AI implementation, though challenges remain.
- AI and SaaS: AI is unlikely to replace SaaS entirely but will make SaaS products smarter.
- VC Model Challenges: AI investment challenges the VC model, potentially giving rise to massive AI funds or ultra-lean AI startups.
- Consumer Applications: Generative AI is expected to revitalize consumer apps and create new giants.
- AI and Blockchain: An interesting direction, but its actual application remains to be seen.
2. Kimi Chat (891 words)
- Structured vs. Unstructured Data: The author notes that currently, unstructured data (ML/AI) receives more attention than structured data (Modern Data Stack).
- Is the Modern Data Stack Dead?: The author believes it faces challenges, including costs and competition from Generative AI.
- Infrastructure Consolidation: Predicts more consolidation and startup failures; big companies will strengthen platform features.
- Databricks vs. Snowflake: Snowflake's growth is slowing, while Databricks shows strong momentum as a key player in Generative AI.
- BI and Generative AI in 2024: Transformation in BI is slow, but Generative AI could deeply impact data analysis.
- The Rise of the Modern AI Stack: Unstructured data infrastructure is having its moment, with AI infrastructure companies growing rapidly.
- Hype Cycle Stage: The current cycle is intense, with the possibility of a bubble.
- Experiment vs. Reality: In 2023, some enterprises invested in Proofs of Concept rather than solving actual business problems.
- LLM Companies: They may not be fully commoditized and might become the new cloud service providers.
- LLMs, SLMs, and Hybrid Futures: Small and specialized models are rising; enterprises will adopt hybrid architectures.
- Is Traditional AI Dead?: Traditional AI and Generative AI are complementary, handling different data types and use cases.
- Thin Wrapper vs. Full Stack: Companies need to become "full-stack" solution providers for specific problems.
- 2024 Areas to Watch: AI agents and Edge AI are exciting fields.
- AGI vs. Plateau: Generative AI progress might slow down rather than accelerating linearly toward AGI.
- GPU Wars and NVIDIA: NVIDIA dominates the market, but future competition and market shifts may pose challenges.
- Open Source AI: The surge in models might lead to market self-adjustment.
- The Actual Cost of AI: Economics are evolving; falling costs are good for users, but construction costs for providers remain high.
- Big Tech Politics: Whether Microsoft has already won the AI race.
- OpenAI Fever: OpenAI's valuation and growth are fascinating, but there are questions about its relationship with Microsoft.
- Year of Enterprise AI: Interest is high, but key implementation issues need resolution.
- Will AI Kill SaaS?: AI will likely empower SaaS products rather than replace them.
- Will AI Kill VC?: AI might change the scale and model of venture capital.
- Consumer Market Revival: The consumer market is looking for the next growth engine; GenAI might be the answer.
- AI and Blockchain: The combination might create new decentralized networks and opportunities.
3. Gemini 1.5 Pro (approx. 1200 words)
- Structured vs. Unstructured Data: Unstructured data is currently receiving significantly more focus than structured data.
- Is the Modern Data Stack Dead?: It faces high costs and declining attention; consolidation and platformization are likely.
- Infrastructure Growth: Many startups will pivot to "AI Infrastructure," with larger companies strengthening their platform dominance.
- Databricks vs. Snowflake: Snowflake's growth is slowing while Databricks is gaining momentum; Microsoft Fabric adds new variables.
- BI and Data Analysis: BI changes slowly, but GenAI has potential to transform analysis if challenges are overcome.
- Modern AI Stack: Faces similar challenges to the modern data stack regarding market size and competition.
- AI Hype Cycle: Currently intense; a correction may occur, so watch for market risks.
- 2023 Reality Check: Successes exist, but there are concerns about real demand and high churn in consumer apps.
- LLM Companies: High business opportunity despite lack of clear model performance dominance, similar to cloud providers.
- Hybrid Future: Specialized and small models are on the rise; hybrid architectures will satisfy specific needs.
- Traditional AI: Remains complementary to GenAI for different use cases.
- Thin Wrappers to Full Stack: Differentiation comes from focusing on specific problems and deep model-level integration.
- 2024 Focus: AI Agents are the "Holy Grail" of automation but are in early stages; Edge AI is also key.
- AGI vs. Plateau: Huge business opportunities exist even if the speed of progress slows down.
- GPU Wars: NVIDIA may face short-term pressure, but long-term prospects for AI chips remain bright.
- Open Source AI: Rapid development but with inconsistent quality across projects.
- AI Costs: Prices are falling, yet construction and service costs remain high.
- Big Tech and Microsoft: Microsoft is aggressive, but Google, Amazon, and Meta are competing heavily.
- OpenAI Ambition: Ambitious goals but execution challenges; its relationship with Microsoft may evolve.
- Enterprise AI Landing: High interest, but issues like ROI, talent, and model reliability need solving.
- AI vs. SaaS: AI is expected to strengthen SaaS rather than kill it.
- AI vs. VC: May require larger funds or conversely reduce the need for traditional VC for lean startups.
- Consumer Market: GenAI could drive growth in search, companions, hardware, and personalized entertainment.
- AI and Blockchain: Potential exists, but beware of industry hype and fraud.
Preliminary Conclusion: Gemini 1.5 seems to provide the longest and best answer, with the best structure and Markdown output.
I then used Claude3 Opus and Gemini 1.5 to compare the three models' answers. Interestingly, Claude3 also agreed that Gemini 1.5 performed better.
Comparison of Model Viewpoints (Summarized by Claude3)
| Topic | Claude3 | Kimi Chat | Gemini 1.5 Pro |
|---|---|---|---|
| BI & GenAI in 2024 | Slow BI change, AI may bring reform | Slow BI change, AI may affect analysis | More specific: AI can change analysis, but challenges exist |
| LLM Company Outlook | Not explicitly mentioned | Won't be commoditized, will be like cloud providers | Massive business opportunities despite low model moat |
| OpenAI Analysis | Not much involved | Valuation, growth, and internal dynamics are key | Ambitious but execution is hard; MS relationship may change |
| Enterprise AI Landing | Promising but faces many challenges | Interest exists, but key problems are unsolved | Lists specific problems to be solved beyond interest |
| AI & Blockchain | A direction, but application is TBD | Possible new decentralized networks | Warns about hype and fraud |
Detailed Comparison Table (Summarized by Gemini 1.5 Pro)
| Subject | Claude3 | Kimi Chat | Gemini 1.5 Pro | Similarities/Differences |
|---|---|---|---|---|
| Structured vs. Unstructured | Unstructured gets more focus | Unstructured gets more focus | Unstructured gets more focus | All three agree |
| Is MDS Dead? | Facing pressure, may be replaced | Facing challenges | Facing cost/attention decline | Claude3 is aggressive; others see challenges |
| Infra Consolidation | Big get bigger | Platforms strengthen | Platformization | All three agree |
| Databricks vs. Snowflake | MS Fabric threat | Snowflake slowing | Snowflake slow, MS Fabric threat | Consistent; Kimi missed MS Fabric |
| AI Infra Startups | Need product pivot | Fast growth | Will pivot to AI Infra | Claude3 notes difficulty; others optimistic |
| AI Hype Cycle | Need to consider stage | Intense, bubble exists | Intense, potential correction | All see intensity; Claude/Gemini more cautious |
| GPU Wars | NVIDIA valuation is a topic | NVIDIA faces challenges | Short-term pressure, long-term bullish | All bullish on chips; views on NVIDIA vary |
| AI Cost | Decline is a boon | Good for users | Good for users, high build cost | All see cost decline; Gemini notes providers |
What are my views on these questions?
Here are some of my own insights:
- Structured vs. Unstructured Data: Unstructured data is clearly more important now, but interestingly, AI actually doesn't understand structured data very well. This is a key area for future improvement.
- Data Stack and Infrastructure: Having tried almost every data stack over the past five years, I agree that there is only room for big companies in the future.
- Databricks/Snowflake, MaaS vs. SaaS: MaaS is undoubtedly the biggest growth point. As I said yesterday: Cloud and Model are both indispensable; I am very optimistic about the Databricks model.
- Hype Cycle and GPUs: I believe we are at a local peak of the hype cycle, and expectations for NVIDIA are extremely optimistic.
- Application Implementation and LLM Commoditization: A standalone LLM is hard to commoditize as a good product, but a model bundled with Cloud can be. The hidden barrier to AI implementation is high; it's a multi-disciplinary engineering problem.
- Edge AI, Consumer Market, and AI Cost: AI cost mainly refers to inference cost. It will drop rapidly, but it's currently very high. Real consideration of AI cost implies existing usage scenarios.
- Open Source: It's a cycle where open source catches up after closed source upgrades, then closed source adopts new ideas from open source and upgrades again. In this gap, data is actually more important than computing power.