Recent trends pairing caviar with chicken nuggets and hot dogs had already created something of an identity crisis for the snooty food. Now, some are bingeing it on the couch, while others are serving it to toddlers.

Recent releases from OpenAI and Anthropic lead other models by a wide margin on Terminal-Bench-Science. GPT-6 Astra and Claude Opus 5.5 have a ~20-point lead over Fable 5.1 in our testing. The best-performing model we’ve evaluated so far outside of these labs is Qwen3.8 Max (0902) at 12%. Within the same model families, the most recent GPT and Claude model releases made large gains. Comparing max effort, GPT-6 Sol and Opus 5.5 both saw performance improvements on the dataset while also reducing cost per task compared to GPT-5.6 Sol and Opus 5.

Recent security incidents show frontier AI labs may be moving too fast. Replit Co-Founder & CEO @amasad warns that labs must prioritize system safety before risking core internet infrastructure: "I think the responsible thing for them to do is to pace and slow down, because otherwise they're going to be criminally liable for hacking core internet infrastructure." #AI# #Cybersecurity# #TheInformation#

Recently had a call with a $50M CEO who thought that AI "couldn't do anything for him". Meanwhile this same guy said he's spending 6-8 hours a day on email. Realistically his time is probably worth $5K/hour (maybe more). So if I charged him $10,000 to build an email triage skill that took me 5 minutes to build it would be such a laughably good deal for him he'd be crazy not to take it. The skill would read every incoming message and drop it into one of 5 buckets: -needs his decision -needs a reply -needs awareness only -can be delegated -can be ignored He told me that roughly 90% of what hits an his inbox is FYI. Zero action required. This skill would surface the 10% and draft the responses (never auto send). That's 4 to 5 hours a day handed back to someone whose time is worth $5,000 an hour or more. Thinking through this as I write this out I should probably charge double.

Recent research has highlighted the promise of scaling memory embeddings in LLM training. While Engram and STEM index memory by token identity or local n-grams, can we design more flexible memory routing that captures how each token’s meaning changes with context? This paper introduces Mixture-of-Memory Embeddings (MoME), which uses a learned router to sparsely select among multiple memory slots for each token. The architecture outperforms strong memory baselines across three model families. Interestingly, both qualitative and quantitative analyses show that the learned routers’ activations correlate with the context-dependent senses of polysemous words. The authors also train a sub-billion-parameter model that achieves competitive CORE-22 performance against similarly sized base models, including Qwen3-0.6B and Llama 3.2-1B. Code and pretrained models are publicly available!

Recent Q2 earnings strength at major U.S. banks was disproportionately driven by trading and investment banking, while recurring business growth lagged, Avi Gilburt @AviGilburt argues. He asks how normalized earnings may look when current tailwinds fade.

Recent moments AI was more annoying than helpful: - Instinct sending me a captcha code to do myself - Grok Bot taking way too long to do something seemingly simple (20 mins to look up some flights) - Hermes got setup to make phone calls with Retell and completely failed by confusing a japanese retail store staff further by asking her why she called lol - Instinct being overly aggressive with scanning my emails unprompted - Grok Bot unable to self install Retell and Twilio, Instinct got farther here but… - Instinct setting up accounts for me then me having to manually reset the password myself + add it into the vault when it can’t get past verification - putting all three (Hermes, Grok Bot, Instinct) into a slack channel and then redoing my calendar events because they got mixed up on timezones - Instinct being over confident in it’s answers when all it did was stop at the first listicle on top of google and gave me the first choice as “best” Today is the worst AI will ever be. I’ve been pushing the limits on experimentation across different models and harnesses daily. It’s a glimpse into what will be possible but there’s still wide gaps for it to be a full on VA or executive assistant