It would depend on exactly how the records were obtained. If hypothetically she paid a nurse $100 for it (and had done so many times in the past), that could be criminally liable. In practice, no-one would prosecute over a single x-ray. But when a reporter gets legally protected medical information "somehow", an insinuation that a legal violation took place at some point to effect that is not unreasonable.
If you induce someone to violate HIPAA who is covered by it (like say a nurse at a hospital), you can be criminally liable. There is no carve-out for journalists. BOTH the person who gave the record and the person who induced them to give it could be liable (not in the same way, possibly). In any case, you seemed to think there was a bright line rule of some sort, that "At one point it accuses Lim of "violating HIPAA", which is not a thing† (HIPAA constrains covered entities, not reporters)." when in fact you can be criminally liable for inducement/conspiracy etc if you induce someone who is covered to give you those records, under https://www.law.cornell.edu/uscode/text/42/1320d-6
Here is another similar case of a non-medical person violating HIPAA.
A. That's how I read it too. B. You can be criminally liable for HIPAA violations, if you induce someone covered by them to violate them. See for example https://www.justice.gov/usao-nj/media/1254226/dl (indictment of KEITH RITSON)
"COUNT 2
(Conspiracy to Wrongfully Obtain and Disclose
Individually Identifiable Health Information)
19. Paragraphs 1-3 and 5-18 of Count 1 of this Superseding Information are
hereby realleged and incorporated as though set forth in full herein.
20. At all times relevant to this Superseding Information:
a. The Health Insurance Portability and Accountability Act of 1996
(“HIPAA”) protects individually identifiable health information from wrongful
disclosure or obtainment and seeks to set national standards to maintain patient
confidentiality.
b. In connection with HIPAA, the United States Department of
Health and Human Services enacted regulations to safeguard the privacy of patients’
medical records and limit circumstances in which individually identifiable health
information or protected health information can be used or disclosed. The HIPAA law
and privacy regulations apply to, among others, health care providers, such as medical
doctors, who transmit health information in connection with a transaction covered by
the law and privacy regulations.
c. Frank Alario, who is listed as a co-conspirator with respect to
Count 2 of this Superseding Information but not as a defendant herein, was a health
care provider and a covered entity under the HIPAA law and privacy regulations.
21. From in or about August 2014 through in or about February 2016, in the
District of New Jersey, and elsewhere, defendant
KEITH RITSON
did knowingly and intentionally conspire and agree with Frank Alario and others to
commit offenses against the United States, that is, to knowingly and without
authorization obtain individually identifiable health information and protected health
information to another person, and to knowingly and without authorization disclose
individually identifiable health information and protected health information
maintained by a covered entity relating to individuals, contrary to Title 42, United
States Code, Section 1320d-6."
Because Cerebras handles large models poorly due to latency/bandwidth issues to main memory. See https://openai.com/index/introducing-gpt-5-3-codex-spark/ where its performance is significantly below that of the regular Codex 5.3, and can only handle a 128k text context window. For some use cases its great, but most would rather use a better, slower model.
In the future, they plan hybrid implementations, to be able to serve large models better, e.g.
"AWS. We signed a binding term sheet with Amazon Web Services for AWS to become the first hyperscaler to deploy Cerebras systems in its data centers. Deployment in AWS data centers will require us to meet strict standards for performance, scale, and reliability.Pursuant to the term sheet, we will create a co-designed, disaggregated inference-serving solution that will integrate AWS Trainium3 chips with Cerebras CS-3 systems, connected via high-bandwidth networking, to partition inference workloads across Trainium3 and CS-3. Each system will perform the type of computation at which it most excels. The approach is expected to deliver 5 times more token throughput in the same hardware footprint, at up to 15 times faster speeds compared to leading GPU-based solutions as benchmarked on leading open-source models."