So this new GPT-4o model (how it's written -- but she's saying "4-oh") update is really about ease of use BUT it has better input for voice, visual, etc. and less latency. And now free users aren't stuck with 3.5. FREE USERS now have access to GPTs and GPT store, greatly widening the market for GPTs. PAID USERS get 5x capacity limit. GPT-4o is now available in API -- says 2x faste and 50 percent cheaper than GPT-turbo.
Rebecca Nagel’s Post
More Relevant Posts
-
the year is not over yet, I bet we'll get more advances in LLMs before the year ends. Mixtral 8x7B, an open weight mixture of experts model. Mixtral matches or outperforms Llama 2 70B and GPT3.5 on most benchmarks, and has the inference speed of a 12B dense model. It supports a context length of 32k tokens. Imagine all the products now possible when inference is 100% free and your data stays on your device. https://lnkd.in/dT7HSMJJ
-
Solution Director at Orange Business Services | Head of CVP Europe | Building the future of CX | 🌍 Ambassador of Digital Transformation | Trusted Advisor | Customer Obsessed | Mentor | Innovation Focus.
GPT-4 Turbo with 128K context, knowledge of world events up to April 2023. GPT-4 Turbo with lower prices, the new Assistants API, GPT-4 Turbo with Vision, DALL·E 3 API, and more.
-
Very accurate. Also true wrt fine tuning. There are good reasons to fine tune! But I highly recommend you first use GPT-4, experiment with prompts, and bring in some custom knowledge in Custom GPTs. Most will be shocked (and satisfied) by how far that gets you in an enterprise context. H/T Ethan Mollick
-
-
Revenue Assurance|Margin Assurance|Data Science|Advanced Analytics|Big Data|Data Evangelist|Machine Learning|AI|GenAI|Tech Enthusiast
Developers can run inference on Llama 3.1 405B on their own infrastructure at roughly 50% the cost of using closed models like GPT-4o, for both user-facing and offline inference tasks. #OpenSourceAI
-
Massive GPT3.5 & 4 improvements from (possibly longest?) context length to finetuning, pricing, speed and supporting multi-modal inputs & outputs (GPT4V for vision, Whisper for text<>speech), but also: - reproducibility with seed parameter - copyright shield; similar to what MS & Google has launched earlier on copyright commitments - custom models (more custom than finetuning but very limited, complex and expensive one!); almost like a whitegloving process by the OpenAI team - GPT Marketplace - GPT Builder (reminds me of Telegram BotFather!) i.e. a GPT that helps you build more GPTs - Assistants API: offloads a lot of the heavy lifting such as embeddings, chunking & search algorithms, adding new functions And continued partnership between OpenAI & Microsoft!
OpenAI announced their new GPT-4 Turbo model today! With significant improvements over GPT-4 · Lower pricing (2.75x cheaper) · Longer context window (up to 128K tokens) · More control with improved function calling, JSON mode and reproducible outputs · Better knowledge (up to April 2023) · New modalities (vision and speech) · Fine-tuning and customization programs · 2x higher rate limits https://lnkd.in/dZb3QsAD
-
-
Been experimenting a little with the custom GPTs. If you have certain topics you find yourself asking a lot of questions regularly, it makes sense to have a dedicated assistant. Creating one is very easy and I think just having basic instructions and context ready when starting a new chat already helps. What really helps is to then use them actively, make notice how it could be / behave better and then refine the instructions over time through trial and error (no user manual this time either). I like this step in GPT evolution very much while being fully aware that these pre-agents will probably soon evolve beyond recognition.
-
-
GPT4-Turbo ✅ new Assistant API for code interpretieur and retrieval ✅ Data up to april 2023 ✅ new Dall-E 3 api ✅ GPT-4 Turbo Vision API ✅ text-to-speech API ( you can choose from 6 preselected voices) ✅ upcoming new model Whisper V3 ✅ 128k context window - equivalent of more than 300 pages of text in a single prompt. ✅ Lower price ✅ 3x cheaper price for input tokens and 2x cheaper price for output tokens compared to GPT-4 LINK:
New models and developer products announced at DevDay
openai.com
-
Building Digital AI Tools for Content Creators using No-Code | Web Designer & Bubble.io Developer
I am currently using the gpt-4o model to generate possible answers & generate a document for an agent. No significant improvements here; however, the new model reduces latency by a few milliseconds. #buildinpublic
-
Prompt Reducer helps you reduce GPT-4 prompts to a few lines, saving tokens and lowering the cost of GPT-4 requests. We utilize a custom prompt to transform GPT-4's input into an encrypted form that it can decrypt later. We know that it's not perfect and it can break sometimes, but we are trying to improve it. If the prompt generated is not working, just try again, it will generate a new one that might work. *So are you trying to keep this prompt as secretive as possible?* Nah, we'll put the prompt we're using in comments and it's on the site too. 👇 https://lnkd.in/gnmBk_m2
Prompt Reducer - Cut Down GPT-4 Token Costs
promptreducer.com
-
GPT4o is now the least expensive, high quality, large parameter LLM on the market. In the video below as you can see I tested GPT-4o with a pretty simple prompt and then I regenerated the completion with the regular GPT-4 response to see the difference. The biggest differences appear to be speed and price. I haven't noticed any particular increase in the accuracy of responses nor any novel capabilities that GPT-4o has that regular GPT-4 doesn't. Our updated MJV LLM Price Comparison sheet is in the comments and you will now see the GPT-4o relative to all the other closed source models out there.
