Немного потестировал Gemini Pro с другими крупнейшими LLM. Никаких сюрпризов, все так же, как и было до этого. Безусловным победителем оказался (сюрприз!) GPT 4 Turbo, за ним идут GPT 4 Bing / Anthropic Claude 2.1. Результат гугловского апгрейда Барда пока не очень впечатляет, в большинстве случаев он отвечает даже хуже Claude, тягаться может разве что с GPT 3.5 Turbo (но тут зависит от задачи, необходимости свежих данных, и это еще пока не Gemini Ultra). Скриншоты с примерами в приложенном архиве:
1. Вопрос про количество бананов:
GPT 3.5 Turbo — ❌ (логическая ошибка)
GPT 4 Turbo — ✅ (отлично)
Gemini Pro / Bard — ❌ (логическая ошибка)
GPT 4 Bing — ❌ (почти ОК, но ошибка в логике)
Anthropic Claude 2.1 — ✅ (отлично)
2. Простая арифметика с порядком операций:
GPT 3.5 Turbo — ❌ (полный провал)
GPT 4 Turbo — ✅ (отлично)
Gemini Pro / Bard — ✅ (отлично)
GPT 4 Bing — ✅ (корявенький вывод)
Anthropic Claude 2.1 — ✅ (отлично)
3. Логика с подсчетом слоновьих ног:
GPT 3.5 Turbo — ✅ (отлично)
GPT 4 Turbo — ✅ (отлично)
Gemini Pro / Bard — ❌ (полный провал, консистентно)
GPT 4 Bing — ✅ (отлично)
Anthropic Claude 2.1 — ✅ (отлично)
4. Шутка Маска с рогозинским батутом:
GPT 3.5 Turbo — ❌ (полный провал)
GPT 4 Turbo — ✅ (отлично)
Gemini Pro / Bard — ✅ (отлично)
GPT 4 Bing — ✅ (отлично)
Anthropic Claude 2.1 — ❌ (полный провал)
5. Вывод слов с заданными параметрами:
GPT 3.5 Turbo — ❌
GPT 4 Turbo — ✅
Gemini Pro / Bard — ❌
GPT 4 Bing — ❌
Anthropic Claude 2.1 — ➖ (почти, но нет)
#llm
1. Вопрос про количество бананов:
GPT 3.5 Turbo — ❌ (логическая ошибка)
GPT 4 Turbo — ✅ (отлично)
Gemini Pro / Bard — ❌ (логическая ошибка)
GPT 4 Bing — ❌ (почти ОК, но ошибка в логике)
Anthropic Claude 2.1 — ✅ (отлично)
2. Простая арифметика с порядком операций:
GPT 3.5 Turbo — ❌ (полный провал)
GPT 4 Turbo — ✅ (отлично)
Gemini Pro / Bard — ✅ (отлично)
GPT 4 Bing — ✅ (корявенький вывод)
Anthropic Claude 2.1 — ✅ (отлично)
3. Логика с подсчетом слоновьих ног:
GPT 3.5 Turbo — ✅ (отлично)
GPT 4 Turbo — ✅ (отлично)
Gemini Pro / Bard — ❌ (полный провал, консистентно)
GPT 4 Bing — ✅ (отлично)
Anthropic Claude 2.1 — ✅ (отлично)
4. Шутка Маска с рогозинским батутом:
GPT 3.5 Turbo — ❌ (полный провал)
GPT 4 Turbo — ✅ (отлично)
Gemini Pro / Bard — ✅ (отлично)
GPT 4 Bing — ✅ (отлично)
Anthropic Claude 2.1 — ❌ (полный провал)
5. Вывод слов с заданными параметрами:
GPT 3.5 Turbo — ❌
GPT 4 Turbo — ✅
Gemini Pro / Bard — ❌
GPT 4 Bing — ❌
Anthropic Claude 2.1 — ➖ (почти, но нет)
#llm
👍2
OpenAI cofounder Ilya Sutskever has become invisible at the company, with his future uncertain, insiders say
The chief scientist and co-founder behind some of OpenAI's biggest breakthroughs in generative AI, who played a key role in the shocking November board ouster of CEO and co-founder Altman, has not been seen at the company's San Francisco offices this week, according to two people familiar with the company.
His position has yet to be officially addressed at OpenAI and there are signs of ongoing tumult.
Sutskever has also hired his own lawyer.
... Sutskever is a "very deep" person, emotionally and intellectually, one of the people familiar said. He often recommends OpenAI employees read The Gulag Archipelago, a nearly 700-page non-fiction book about the system of forced labor in the Soviet Union...
... One Microsoft insider familiar with Altman, Sutskever, and Brockman doesn't believe the three can ever work well together again, particularly Sutskever and Brockman. In Silicon Valley, founders turning on one another is considered sacrilege.
Some of OpenAI's engineers who are loyal to Altman and Brockman may also find it difficult to work with Sutskever because of his role in the ouster, a former employee said. "Once trust is broken," the former staffer said, "it cannot be regained."
Источник: Business Insider
Версия в вебархиве, без пейволла
The chief scientist and co-founder behind some of OpenAI's biggest breakthroughs in generative AI, who played a key role in the shocking November board ouster of CEO and co-founder Altman, has not been seen at the company's San Francisco offices this week, according to two people familiar with the company.
His position has yet to be officially addressed at OpenAI and there are signs of ongoing tumult.
Sutskever has also hired his own lawyer.
... Sutskever is a "very deep" person, emotionally and intellectually, one of the people familiar said. He often recommends OpenAI employees read The Gulag Archipelago, a nearly 700-page non-fiction book about the system of forced labor in the Soviet Union...
... One Microsoft insider familiar with Altman, Sutskever, and Brockman doesn't believe the three can ever work well together again, particularly Sutskever and Brockman. In Silicon Valley, founders turning on one another is considered sacrilege.
Some of OpenAI's engineers who are loyal to Altman and Brockman may also find it difficult to work with Sutskever because of his role in the ouster, a former employee said. "Once trust is broken," the former staffer said, "it cannot be regained."
Источник: Business Insider
Версия в вебархиве, без пейволла
Business Insider
OpenAI cofounder Ilya Sutskever has become invisible at the company, with his future uncertain, insiders say
The OpenAI co-founder was an integral part of what became ChatGPT. Insiders say his role in Sam Altman's ouster is showing ongoing damage.
Анонимный телеграм-канал
Особенно интересна часть про автоматизацию: https://www.youtube.com/watch?v=FbCbXGgkhjE
А теперь для сравнения: примерно в то же время (1982 год) Sony выпустила первый проигрыватель компакт-дисков CDP-101.