Did 5.6 Sol Just Close The Fable Gap?
The Daily AI Show
1 DAYS AGO
Did 5.6 Sol Just Close The Fable Gap?
Did 5.6 Sol Just Close The Fable Gap?
The Daily AI Show
1 DAYS AGO
In this episode of the Daily AI Show, the hosts dive into OpenAI's latest ChatGPT Work rollout, exploring its new desktop experience and how it integrates with Codex, browser control, and mobile workflows. They compare GPT-5.6 Sol and Terra against Fable, focusing on coding performance, agentic capabilities, and cost efficiency. The discussion also covers the practical implications of computer use for local automation, the differences between ChatGPT Work and Claude Co Work, and the potential for AI agents to operate other AI tools. News highlights include Fiji Simo's departure from OpenAI, AMD's compact AI PC, a Brown University AI cheating incident, and Nvidia's NemoClaw release.
The hosts analyze ChatGPT Work's three-way split into browser, desktop, and mobile, noting desktop's local file access and mobile's command center role. They find GPT-5.6 Sol outperforms Claude in client projects by asking better questions and generating superior HTML, though desktop app sync issues persist. Codex's computer use is praised for smooth automation of tasks like AWS configuration and grocery lists, with minimal token usage. The episode contrasts ChatGPT Work's browser control with Claude Co Work's Playwright reliability, and highlights cross-platform synergy by using Codex to operate Claude. News covers Fiji Simo's health-related departure from OpenAI, AMD's $1,499 AI PC with 128GB shared memory, and a Brown University study showing AI use as an answer engine dropped test scores from 96% to 48%. The hosts advocate for AI as a learning tool, not a crutch, and call for regulation of AI use by minors. Nvidia's NemoClaw achieves state-of-the-art agentic benchmarks at $4.48 per task, challenging competitors in enterprise AI.
00:00
00:00
Major ChatGPT announcement and GPT-5.6 Sol and Terra details.
00:44
00:44
Live translation demo had a comedic error
03:50
03:50
GPT-5.6 is significantly cheaper per task than Fable.
10:30
10:30
ChatGPT's naming is like the Dave Matthews Band's accidental naming
12:11
12:11
Usage limits now visible with reset times
17:26
17:26
Local file access is the key advantage.
18:53
18:53
Desktop handles local files, mobile acts as a command center
21:24
21:24
Playwright's reliability in Claude Co Work
22:47
22:47
ChatGPT Work independently controls a browser with its own mouse.
28:01
28:01
Codex's smooth interface and minimal token usage
31:38
31:38
Autonomous desktop automation is here.
35:24
35:24
Store API keys in .env files, not in chat.
39:06
39:06
Using Codex in ChatGPT to operate Claude desktop app
40:44
40:44
AI merely generates plausible responses without true understanding.
44:46
44:46
Fiji Simo steps down from OpenAI due to a worsening neuroimmune condition
48:05
48:05
AMD's new AI PC with 128GB shared memory for $1500 is compared to Apple's high-end Macs.
52:00
52:00
Struggle is essential for cognitive development.
53:57
53:57
Learning should happen where knowledge is applied.
56:34
56:34
Nemotron 3 achieves state-of-the-art at $4.48 per task
1:00:00
1:00:00
GPT for personal, Claude for work.
