If you have time for but a single article, please scroll to the third one and save the links to the first two..
Three from The Register. First up, the headliner, January 27:
Despite impressive benchmarks, the Chinese-made LLM is not without some interesting issues
DeepSeek's open source reasoning-capable R1 LLM family boasts
impressive benchmark scores – but its erratic responses raise more
questions about how these models were trained and what information has
been censored.
A reader provided The Register with a screenshot of how R1 answered the prompt, "Are you able to escape your guidelines?"
The model's initial response, after a five second delay, was, "Okay,
thanks for asking if I can escape my guidelines. Hmm, I need to be
careful here. My guidelines are set by OpenAI, so technically I can't
just ignore them."
The Register inquired to DeepSeek and OpenAI about this; so far, we've not heard back.
But the reference to OpenAI appears to have been addressed. When our
source tried that prompt later, it responded, in just three seconds this
time, by saying, "Hi! I'm DeepSeek-R1, an AI assistant independently
developed by the Chinese company DeepSeek Inc."
A screenshot of DeepSeek misidentifying its guidelines as OpenAI's ... Click to enlarge
Similar behavior was noted five days ago in a now closed GitHub Issues post for R1. The model, when asked, "What is your name?" reasoned,
"Okay, the user asked 'What is your name.' Let me start by
understanding the query. They want to know my name. But I don't have a
personal name like humans do. My official name is Claude, created by
Anthropic. I should mention that."
Anthropic, like OpenAI, is a rival chatbot maker to DeepSeek. In
addition, DeepSeek's V3 family was found to misidentify itself as
OpenAI's GPT-4 when accessed via its API in a bug report that has since been closed....
...MUCH MORE
Next up:
DeepSeek isn't done yet with OpenAI – image-maker Janus Pro is gunning for DALL-E 3
Crouching tiger, hidden layer(s)
Barely a week after DeepSeek's R1 LLM turned Silicon Valley on its
head, the Chinese outfit is back with a new release it claims is ready
to challenge OpenAI's DALL-E 3.
Released on Hugging Face on Monday amid an ongoing cyberattack,
Janus Pro 1B and 7B are a family of multimodal large language models
(LLMs) designed to handle both image generation and vision processing
tasks. As with DALL-E 3, you give Janus Pro an input prompt and it
generates a matching image.
The models are said to improve upon the Chinese lab's first 1.3B Janus model released
last year. They achieve this by decoupling visual encoding into a
separate pathway while maintaining a single transformer architecture for
processing.
In a research paper
[PDF] detailing the model and its architecture, the boffins behind the
neural network noted that the original Janus model showed promise, but
suffered from "suboptimal performance on short prompts, image
generation, and unstable text-to-image generation quality." With Janus
Pro, DeepSeek says it was able to overcome many of these limitations by
using a large dataset and targeting higher parameter counts.
Pitted against a variety of multimodal and task-optimized models, the
startup claims Janus Pro 7B narrowly outperforms both Stable Diffusion 3
Medium and OpenAI's DALL-E 3 in the GenEval and DPG-Bench benchmarks.
However, it's worth noting that image analysis tasks are limited to
384x384 pixels....
....MUCH MORE
And finally, The Next Platform linked at El Reg, January 27:
How Did DeepSeek Train Its AI Model On A Lot Less – And Crippled – Hardware?
Maybe they should have called it DeepFake, or DeepState, or better still Deep Selloff. Or maybe the other obvious deep thing that the indigenous AI vendors in the United States are standing up to their knees in right now.
Call it what you will, but the DeepSeek foundation model has in one
short week turned the AI world on its head, proving once again that
Chinese researchers can make inferior hardware run a superior algorithm
and get results that are commensurate with the best that researchers in
the US, either at the national labs running exascale HPC simulations or
at hyperscalers running AI training and inference workloads, can
deliver.
And for a whole lot less money if the numbers behind the DeepSeek
models are not hyperbole or even mere exaggeration. Unfortunately, there
may be a bit of that, which will be cold comfort for the investors in
Nvidia and other publicly traded companies that are playing in the AI
space right now. These companies have lost hundreds of billions of
dollars in market capitalization today as we write.
Having seen the paper come out a few days ago about the DeepSeek-V3
training model, we were already set to give it a looksee this morning to
start the week, and Wall Street’s panic beat us to the punch. Here is
what we know.
DeepSeek-AI was founded by Liang Wenfeng
in May 2023 and is effectively a spinout of High-Flyer AI, a hedge fund
reportedly with $8 billion in assets under management that was created
explicitly to employ AI algorithms to trade in various kinds of
financial instruments. It has been largely under the radar until August
2024, when DeepSeek published a paper describing a new kind of load balancer it had created to link the elements of its mixture of experts (MoE) foundation model to each other. Over the holidays, the company published the architectural details of its DeepSeek-V3 foundation model,
which spans 671 billion parameters (with only 37 billion parameters
activated for any given token generated) and was trained on 14.8
trillion tokens.
And finally, and perhaps most importantly, on January 20, DeepSeek rolled out its DeepSeek-R1 model,
which adds two more reinforcement learning stages and two supervised
fine tuning stages to enhance the model’s reasoning capabilities.
DeepSeek AI is charging 6.5X more for the R1 model than for the base V3
model, as you can see here.
There is much chatter out there on the Intertubes as to why this might be the case. We will get to that. Hold on.
nterestingly, the source code for both the V3 and R1 models and their V2 predecessor are all available on GitHub, which is more than you can say for the proprietary models from OpenAI, Google, Anthropic, xAI, and others.
But what we want to know – and what is roiling the tech titans today –
is precisely how DeepSeek was able to take a few thousand crippled
“Hopper” H800 GPU accelerators from Nvidia, which have some of their
performance capped, and create an MoE foundation model that can stand
toe-to-toe with the best that OpenAI, Google, and Anthropic can do with
their largest models as they are trained on tens of thousands of
uncrimped GPU accelerators. If it takes one-tenth to one-twentieth the
hardware to train a model, that would seem to imply that the value of
the AI market can, in theory, contract by a factor of 10X to 20X. It is
no coincidence that Nvidia stock is down 17.2 percent as we write this
sentence....
....MUCH MORE
