weightedvision
@weightedvision.bsky.social
Exploring LLMs, GenAI, and digital delusions. One of us is hallucinating, and it’s not always the model.
Some early Gemini 2.5 Pro (w/ Deep Think) benchmarks 🚀
June 22, 2025 at 9:44 PM
Some early Gemini 2.5 Pro (w/ Deep Think) benchmarks 🚀
1/
Prompt engineering isn’t just aesthetics—it changes outcomes.
I ran a quick benchmark testing how GPT-4o and Claude S4 solve a rebus puzzle using weak vs engineered prompts. Same task. Same models. Very different results.
Prompt engineering isn’t just aesthetics—it changes outcomes.
I ran a quick benchmark testing how GPT-4o and Claude S4 solve a rebus puzzle using weak vs engineered prompts. Same task. Same models. Very different results.
June 18, 2025 at 12:55 PM
1/
Prompt engineering isn’t just aesthetics—it changes outcomes.
I ran a quick benchmark testing how GPT-4o and Claude S4 solve a rebus puzzle using weak vs engineered prompts. Same task. Same models. Very different results.
Prompt engineering isn’t just aesthetics—it changes outcomes.
I ran a quick benchmark testing how GPT-4o and Claude S4 solve a rebus puzzle using weak vs engineered prompts. Same task. Same models. Very different results.
The progress of Gemini over the last year 🚀
Gemini 2.5 Pro is benchmarking like it's trying to speedrun the Turing test. If this is where we are now… I’m afraid to ask what next year looks like 👀
Gemini 2.5 Pro is benchmarking like it's trying to speedrun the Turing test. If this is where we are now… I’m afraid to ask what next year looks like 👀
June 18, 2025 at 10:20 AM
The progress of Gemini over the last year 🚀
Gemini 2.5 Pro is benchmarking like it's trying to speedrun the Turing test. If this is where we are now… I’m afraid to ask what next year looks like 👀
Gemini 2.5 Pro is benchmarking like it's trying to speedrun the Turing test. If this is where we are now… I’m afraid to ask what next year looks like 👀
1/
Google just expanded its Gemini 2.5 lineup, officially launching Gemini 2.5 Pro and Flash as generally available models, and introducing a new Flash‑Lite variant now in public preview.
The Gemini stack is starting to look more complete, and more competitive.
Google just expanded its Gemini 2.5 lineup, officially launching Gemini 2.5 Pro and Flash as generally available models, and introducing a new Flash‑Lite variant now in public preview.
The Gemini stack is starting to look more complete, and more competitive.
June 17, 2025 at 9:34 PM
1/
Google just expanded its Gemini 2.5 lineup, officially launching Gemini 2.5 Pro and Flash as generally available models, and introducing a new Flash‑Lite variant now in public preview.
The Gemini stack is starting to look more complete, and more competitive.
Google just expanded its Gemini 2.5 lineup, officially launching Gemini 2.5 Pro and Flash as generally available models, and introducing a new Flash‑Lite variant now in public preview.
The Gemini stack is starting to look more complete, and more competitive.
So now I can summon cursed AI art in WhatsApp right between “ok 👍” and “seen 2:14 PM”... Thanks, ChatGPT.
June 17, 2025 at 5:23 PM
So now I can summon cursed AI art in WhatsApp right between “ok 👍” and “seen 2:14 PM”... Thanks, ChatGPT.
Cursor just soft-launched a $200/mo “Ultra” plan: Unlimited everything, 20x usage across OpenAI, Claude, Gemini, plus early access to features. Basically God mode for your IDE.
June 17, 2025 at 5:21 PM
Cursor just soft-launched a $200/mo “Ultra” plan: Unlimited everything, 20x usage across OpenAI, Claude, Gemini, plus early access to features. Basically God mode for your IDE.