A paper on arXiv by a group of authors working for Facebook, which is now calling itself “Meta”, “OPT: Open Pre-trained Transformer Language Models” reports the development and release of their independent large language model, similar in structure and goals to OpenAI’s GPT-3. Here is the abstract.
Large language models, which are often trained for hundreds of thousands of compute days, have shown remarkable capabilities for zero- and few-shot learning. Given their computational cost, these models are difficult to replicate without significant capital. For the few that are available through APIs, no access is granted to the full model weights, making them difficult to study. We present Open Pre-trained Transformers (OPT), a suite of decoder-only pre-trained transformers ranging from 125M to 175B parameters, which we aim to fully and responsibly share with interested researchers. We show that OPT-175B is comparable to GPT-3, while requiring only 1/7th the carbon footprint to develop. We are also releasing our logbook detailing the infrastructure challenges we faced, along with code for experimenting with all of the released models.
We already get that “uncanny valley” feeling similar to that induced by seeing a picture of Mark Zuckerberg and his minions:
upon reading the bit about “requiring only 1/7th the carbon footprint to develop”, which dead horse (how much carbon is one of those?) is beaten savagely in the full paper:
Furthermore, there exists significant compute and carbon cost to reproduce models of this size. While OPT-175B was developed with an estimated carbon emissions footprint (CO2eq) of 75 tons, GPT-3 was estimated to use 500 tons (Patterson et al., 2021), while Gopher required 380 tons (Rae et al., 2021). These estimates are not universally reported, and the accounting methodologies for these calculations are also not standardized. In addition, model training is only one component of the overall carbon footprint of AI systems; we must also consider experimentation and eventual downstream inference cost, all of which contribute to the growing energy footprint of creating large-scale models (Wu et al., 2022). By releasing our logbook, we hope to highlight the gap between a theoretical carbon cost estimate that assumes no hardware failures or training instabilities, versus one that aims to include the entire LLM development lifecycle. We need to understand the manufacturing (or embodied) carbon of these systems (Gupta et al., 2021) as they grow increasingly more complex, and we hope that our paper can help future work in defining additional factors to consider when measuring the impact of scale on the environment.
But the real fun starts when they get going on “Bias & Toxicity Evaluations” in section 4.
To understand the potential harm of OPT-175B, we evaluate a series of benchmarks related to hate speech detection, stereotype awareness, and toxic content generation. While there may be shortcomings in these benchmarks (Blodgett et al., 2021; Jacobs and Wallach, 2021), these measurements provide a first step towards understanding the limitations of OPT-175B. We compare primarily against GPT-3 Davinci, as these benchmarks were not yet available to be included in Brown et al. (2020).
They then go on to feed their big 175 billion parameter model prompts from a series of benchmarks intended to elicit “hate speech”, “intrasentence level biases”, “stereotypical bias”, “toxic language”, “explicit safety failures” and conclude (section 5):
As shown in Section 4, we also find OPT-175B has a high propensity to generate toxic language and reinforce harmful stereotypes, even when provided with a relatively innocuous prompt (Gehman et al., 2020), and adversarial prompts are trivial to find (Dinan et al., 2021). There has been a great deal of work on mitigations for toxicity and biases (Dathathri et al., 2019; Dinan et al., 2019a; Sheng et al., 2019; Dinan et al., 2020a; Liu et al., 2019a; Krause et al., 2020; Xu et al., 2020; Liang et al., 2021; Dinan et al., 2021; Xu et al., 2021a; Dhamala et al., 2021; Schick et al., 2021; Ouyang et al., 2022). Depending on downstream applications, future uses of OPT-175B may need to employ these or novel mitigation approaches, especially before any real world deployment. Given our primary goal as a replication of GPT-3, we choose not to apply these mitigations in this first release
Consequently, while the code and smaller pre-trained models are posted on GitHub, the full model will only be released “responsibly” to “interested researchers”, particularly in “Responsible AI”.
So, let’s see what happened here. They built this language model, trained it with 180 billion words (800 gigabytes) of data sucked up indiscriminately from sources such as Reddit, HackerNews, Wikipedia, Project Gutenberg, the U.S. Patent Office, etc., etc., and to their horror discovered that when asked common questions, it answered as normal humans being do, not in the Silicon Valley court language spoken by people who look like those in the picture above.
We believe the entire AI community would benefit from working together to develop guidelines for responsible LLMs, and we hope that broad access to these types of models will increase the diversity of voices defining the ethical considerations of such technologies.
Indeed…. Perhaps they could train it on the Newspeak dictionary we hear is being prepared by the Disinformation Governance Board.
