Brought to you by:
Top stories today:
- OpenAI releases fine-tuning on GPT-4o to offer customization
- Phi-3.5 models released: on par with Llama 3.1, GPT 4o-mini
- $13B X buyout loans become longest ever to remain in banksā books
- Defense tech startups raise $2.5B YTD, +19% from 2023
- Global foundry revenue +23% YoY in Q2
0. Data and calendar
All values as of 6 AM ET / 3 AM PT, other than S&P500 and NASDAQ close (4 PM ET / 1 PM PT).
All times are ET.
1. OpenAI releases fine-tuning on GPT-4o to offer customization options
A fine-tuned GPT-4o model (Genie) ranked #1 on the code-solving SWE-bench criterion, hitting 43.8%, the benchmarkās ālargest ever improvementā:
Available now to paid usage tiers.
āLayer safety mitigationsā implemented to avoid misuse.
Cost: $25/1M tokens for fine-tuning, $3.75 input, $15 output.
1M free training tokens per day for every organization through Sep. 23.
Fine-tuning was earlier available on other OpenAI models, including GPT-4o mini.
A fine-tuned GPT-4o ranked #1, but still behind humans, on Distylās text-to-SQL benchmark
Relatedly, OpenAI struck a multiyear deal with CondƩ Nast to expand publisher reach.
Automation and integration that scales processes and delivers insights.
From sale to billing and revenue recognition to general ledger and forecasting, get one system of record.
*Sponsored.
3. Phi-3.5 models released by Microsoft: 4B, open weights, on par with Llama 3.1, GPT 4o-mini
Phi-3.5-MoE 4B beats Llama 3.1 8B on most benchmarks, but still loses to GPT 4o-mini (size not divulged by OpenAI) on all but one:
