Actus IA : la révolution des "petites" IA (LLM vs SLM)

Actus IA : la révolution des "petites" IA (LLM vs SLM)

by Ludo Salenne

Applied Sciences & Engineering AI — Models & Research UYQArtificial intelligence (AI)UYQMMachine learning
🎙 Ludo Salenne 👥 267K 📅 April 26, 2024 ⏱ 29 min 👁 12K 📄 news review 🧭 2026-08-21
Available in: English (current) Français

Keywords

SLMLLMOpenAIMicrosoftApple

Summary

This weekly AI news review covers several major developments. The creator discusses the OpenAI-Moderna partnership, highlighting the use of ChatGPT in healthcare and the associated risks. He then presents the Hugging Face Open Medical LLM Leaderboard, showing that specialized models outperform general ones but with significant variability across medical fields. The video also covers SenseTime’s SenseChat 5.0, which claims to beat GPT-4 Turbo, but the creator notes the comparison is against an older version. Stanford’s AI Index report is summarized, showing the US leading in AI models and private investment, with France in third place for models. The creator tests DuckDuckGo’s new AI chat, which offers privacy-focused access to GPT-3.5. Adobe’s Firefly 3 and VideoGigaGAN are showcased, with a live test of Firefly 3’s image generation. The main theme is the rise of Small Language Models (SLMs), with Apple’s OpenELM and Microsoft’s Phi-3 announced. The video also covers Meta’s Ray-Ban glasses, TikTok’s voice cloning, Grok in Tesla, and Elon Musk’s legal actions against OpenAI.

164 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides a good overview of recent AI news, with a clear focus on the shift towards smaller, more efficient models. The creator’s argumentation is generally sound, but sometimes relies on personal opinion rather than deep analysis. For example, the discussion on the OpenAI-Moderna partnership raises valid concerns about AI reliability in healthcare, but the creator does not delve into the specifics of how Moderna uses the technology. The comparison of SenseChat 5.0 with GPT-4 Turbo is well-handled, noting the version discrepancy. The test of DuckDuckGo AI is practical and informative. The presentation of Stanford’s AI Index is useful, but the creator’s interpretation of the data could be more nuanced.

Scientific Rigor, Source Quality, Title Accuracy

The creator provides links to sources for most claims, which is commendable. The sources include official announcements from OpenAI, Microsoft, Apple, and Meta, as well as articles from reputable tech outlets. However, some claims, such as the performance of SenseChat 5.0, are based on a Reddit post and the company’s own press release, which may be biased. The title accurately reflects the content, focusing on the SLM revolution. The video is a news review, so it does not provide original research, but it does synthesize information from multiple sources.

215 words

Title / Content Match

The title accurately reflects the content, which focuses on the emergence of small language models (SLMs) as a key trend, while also covering other AI news.

Quality & Reliability

7/10

The video is a weekly news review covering multiple AI developments. The creator provides links to sources for most claims, but some interpretations are personal and not deeply verified. The information is generally accurate and up-to-date as of the publication date, but the analysis is at a surface level.

Chapters

Cited Sources

  • OpenAI and Moderna partnership — Discussed at the beginning of the video, highlighting the use of ChatGPT in healthcare.
  • Open Medical LLM Leaderboard — Referenced to show the performance of medical AI models.
  • Stanford AI Index 2024 — Used to discuss the global AI landscape and performance vs. humans.
  • SenseChat 5.0 announcement — Mentioned as the source for the claim that a Chinese AI outperforms GPT-4 Turbo.
  • Reddit discussion on SenseChat 5.0 — Referenced as the source for the benchmark table shown in the video.
  • DuckDuckGo AI Chat — The video tests this new feature.
  • Microsoft Phi-3 announcement — Discussed in the context of the SLM revolution.
  • Apple acquires Datakalab — Reported as part of Apple's AI strategy.
  • Apple's AI acquisitions in 2023 — Provides context on Apple's acquisition strategy.
  • Perplexity funding — Announced that Perplexity became a unicorn.
  • Meta Ray-Ban glasses update — Discussed in the context of AI wearables.
  • TikTok voice cloning leak — Mentioned as a security concern.
  • arXiv paper on VideoGigaGAN — Referenced for the video upscaling AI.

Concurring Sources

  • Stanford AI Index 2024 — Supports the claim that AI outperforms humans in many tasks.
  • Microsoft Phi-3 announcement — Confirms the trend of SLMs.

Dissenting Sources

  • SenseChat 5.0 benchmark claims — The benchmarks are from the company itself and may be biased; the comparison was against an older GPT-4 Turbo version.

External References

Contribution & Novelties

The video provides a timely overview of the shift towards Small Language Models (SLMs), highlighting announcements from Apple and Microsoft. It also offers practical tests of new tools like DuckDuckGo AI and Adobe Firefly 3. The discussion on the Stanford AI Index adds valuable context on the global AI race.

Pour aller plus loin :

86 words

Radar Profile

The radar profile shows a balanced video with high information quantity and moderate technical depth. The reliability is decent, but the analysis is not deeply critical. The video is more of a news roundup than an in-depth technical review.

Reliability 7/10

💬 Sur les 0 commentaires analysés, aucune tendance n'est disponible.